codeg: starejše različice za prenos

This release redesigns how conversations show an agent’s work: a new activity timeline brings thoughts, commands, file changes and sub-agent tasks together, making progress easier to follow and results easier to review.
New
- A new activity timeline presents the agent’s work step by step, with a summary of the step count, elapsed time and failures.
- Tool permission decisions stay with the conversation, so “approved once”, “always allowed” and “declined” records remain available after you reopen it.
- File edits in conversations support inline and side-by-side comparison, with synchronized horizontal scrolling and a file header that opens the file in the workspace.
Improved
- Clicking the Activity header expands or collapses the steps, while clicking its failure indicator takes you straight to the failed step.
- Consecutive file lookups and commands are grouped automatically, with completed command groups folding away while unresolved failures stay open.
- Command rows explain what the command does when the agent provides a description, and their detail panels show the full command or script alongside its output, with separate copy actions.
- Sub-agent tasks show their task, completed steps, current work and latest result, and running tasks remain visible when you collapse the parent reply.
- Message navigation in Session Details uses a more compact layout for sub-agents and changed files, while messages, code blocks and detail panels blend more naturally with workspace background images.
- Long conversations and streaming replies feel smoother, with steps remaining clickable during scrolling and no more repeated jumping near the bottom.
Fixed
- Command results show exit codes more reliably, including codes recorded separately from the output by Pi, Cursor and Antigravity.
- Codex’s context-compaction indicator appears while compaction is running, instead of only after it finishes.
- Expanded thoughts preserve their line breaks and list formatting, and English text, code and diffs display correctly in the Arabic interface.
Note
- Back up codeg’s data folder before updating if you may need to return to 0.39.0, because 0.40.0 updates the database and 0.39.0 cannot start on that updated data, including through codeg-server’s Roll back or reinstalling the older desktop app.

This one lets an agent write your commit messages, lets a command run on your own computer even from a remote folder, and lets codeg-server sign people in through a proxy such as Cloudflare Access instead of its access token.
Alongside that: Recent in the sidebar lists what you used last, on Windows an update waits for Restart to update and no longer stops at "Failed to kill codeg", the Linux AppImage no longer hands its own libraries to agents and the terminal, and a phone that comes back to the page gets the rest of the reply.
New
- An agent can write your commit message: the ✨ button by the commit box, in the Changes tab and the commit window, has the agent of the conversation you're in write one from exactly what will be committed, taking anything you've typed as a hint, and adds no conversation to your list. (#931 requested by @trueai-org, #928 by @Time-w)
- With untracked files around, the ✨ button first asks which of them the message covers, and Enter then commits exactly those, with a line under the commit box saying how many it will take.
- Commit message settings… in the More Git actions menu chooses who writes the message and how: a fixed agent, model and options, instructions it always follows such as the language or format, and committing as soon as it's written, for all folders or just one project.
- Manage commands has a Run on this computer switch, so a command such as opening the remote folder in your local VS Code or Zed over SSH runs on the computer with the desktop app, and Show in all folders keeps it at hand across folders, worktrees and remote connections. (#974, @fengpanhe)
- codeg-server can sign people in through a proxy such as Cloudflare Access: with CODEG_AUTH=proxy-jwt, a browser the proxy has let in no longer asks for the access token, which codeg-cli and scripts that bypass the proxy keep using, as the proxy sign-in guide explains. (#985, @MagTer)
- Grok's custom model card can declare the model's reasoning-effort levels, so the composer offers a reasoning-effort selector for a custom model too. (#993 requested by @HOMEFTW)
Improved
- Recent in the sidebar lists the conversations you used last instead of the ones created last, ten at a time, with each row showing when it was last active, and Chat rows show their last activity too. (#1010, @Flyneen)
- On Windows, a downloaded update waits for you to click Restart to update, as on macOS and Linux, instead of closing codeg the moment the download finishes, and an install that fails says so and can be tried again.
- The commit box in the Changes tab grows with the message, with Shift+Enter for a new line while Enter still commits.
- Clicking a delegation card unfolds the full task the sub-agent was given, and a resumed sub-agent's card shows there why it was resumed.
- The sidebar can be dragged narrower on Windows, Linux and in the browser, since only the macOS app needs room for its window buttons.
- Pi's model selector lists only the models your enabledModels setting allows. (#989, @xyalter)
- ZCode takes pasted and uploaded images and PDFs, and keeps each reply's token usage and time after you reopen a conversation. (#1003, #1005, @jackie-cqz)
- Claude Code moves to claude-agent-acp 0.94.0, and along with newer CodeBuddy, Cursor, Kimi Code, Qoder, ZCode and Command Code, a Claude Code fork sub-agent's card now shows its real report instead of "Fork started".
Fixed
- Updating codeg on Windows no longer stops at "Failed to kill codeg" while Windows or an antivirus is checking codeg.exe.
- In the Linux AppImage, the agents, terminals and other programs codeg starts get your system's own environment instead of the AppImage's, so on Ubuntu 26.04 installing and starting agents, npm and terminal commands no longer fail with "LIBSYSTEMD_254 not found". (#983 reported by @yanvigdev)
- A phone that puts the page in the background mid-reply gets the rest of the reply when it comes back, without reloading the page. (#386 reported by @wuzhaozhongguo)
- Behin

This one adds codeg-cli, a command that runs agents through codeg from a terminal, a script or a CI job and prints the reply as text or JSON, while each run stays a conversation you can open and carry on in codeg.
Alongside that: a browser picks up the server's new release after an upgrade, /deny in a chat channel no longer gives Claude Code a lasting permission, tables and code blocks have one border again with code highlighted in light mode, codeg restarts into the Dock icon you pinned on macOS, and installing or updating on Windows no longer hangs while stopping codeg-mcp.
New
- codeg-cli runs agents through codeg from a terminal, a script or a CI job: codeg-cli -a claude_code -p "Review the latest commit" hands the prompt to the desktop app or codeg-server already running on this computer, or to a codeg-server elsewhere with --url and its token, and prints the reply as text, JSON or a stream of JSON events with documented exit codes, while the run is a conversation in codeg like any other. (#694 requested by @yasin1151)
- --permissions decides how a run with no one watching answers the agent's permission requests: refuse (the default), allow reads only, allow, end the run, or leave them for you in codeg, and it only ever answers for this once, never with an "always" that would write a lasting rule into the agent's settings.
- A run can go on in the background and be collected later with --detach, result and cancel, work in a git worktree of its own with --worktree, or go to several agents or assistants at once, and --output-schema has the agent hand back JSON that matches a schema you give.
- codeg-cli also lists and installs agents, lists your assistants and runs saved automations, and --standalone gives a single command, in a CI job for example, a codeg-server of its own that goes away when the command ends.
- Settings → System → Command-line access is the switch for codeg-cli on this computer and shows a command to try, while in the desktop app Install command-line tool puts codeg-cli on your PATH, and the app's own terminals and agents find its codeg-cli first.
- Conversations started from the command line carry a terminal mark in the sidebar, and Show in Recent gains Command line to list only them.
- codeg-cli comes with the desktop app, the server packages and the Docker image, install.sh and install.ps1 put it next to codeg-server, and from now on codeg-server's in-place update keeps it up to date. The command-line guide covers every command and option.
Improved
- Local images also show in permission requests, plan approvals and a to-do's Result, as they do in replies.
Fixed
- A browser picks up the server's new release after an upgrade, instead of sometimes opening the previous one from its cache, and a tab left open through the upgrade offers to reload. (#980 reported by @wuzhaozhongguo)
- /approve and /deny in a chat channel answer just this once and pick the answer by what it does, not by where it sits in the list: for a call Claude Code flags as risky, /deny could give it a lasting permission while /approve and automatic approval refused it. A request with no one-time answer now waits for you in codeg.
- Tables and code blocks in replies, thinking and Markdown previews have a single border again, with a code block's language, download and copy buttons in a bar above the code, and code in light mode is highlighted again.
- A table opened full screen keeps its exit button on screen.
- On macOS, codeg comes back after an update or a restart in the Dock icon you pinned, instead of as a second icon next to it.
- Installing or updating codeg on Windows no longer hangs at "Stopping codeg-mcp and codeg-computer-helper processes…" on a PC where Windows Management Instrumentation (WMI) doesn't answer, and install.ps1 waits at most 15 seconds for it.
- Two sub-agents delegated at the same moment into a folder codeg hadn't seen yet both start, instead of one of them failing.
Note
- Once 0.38.0 has run, 0.37.0 no longer sta

This one adds TRAE CLI and Command Code as built-in agents, lets a Claude Code session continue on your phone with Remote Control, and has Claude Code suggest your next prompt. The side panel gains Automations, To-dos, Issues and Pull requests tabs and now holds Agent Plan and Message navigation, the model and option pickers under the message box merge into one button, and Claude Code moves up to 0.93.0 along with seven more bundled agents.
Alongside that: a sub-agent waiting on you can be answered right where it was delegated, images in replies show even when they are outside the conversation's folder, a reply that arrives just as a turn ends is no longer lost, switching Cline's model mid-conversation takes effect, saving a Cursor MCP server in codeg keeps Cursor's own settings, and computer use moves to cua-driver 0.36.0.
New
- ByteDance's TRAE CLI is now a built-in agent: codeg runs the traecli you installed, with its mode, model and reasoning effort pickers in the message box, its MCP servers and skills managed from codeg's settings, delegation, and history for the conversations you run in codeg, while its settings page offers Upgrade and an optional personal access token for servers and Docker. (#933 requested by @lwxk)
- Command Code is now a built-in agent: install it from Settings → Agents and sign in once with command-code login or save an API key, and codeg runs it with its mode, model and effort pickers in the message box, its MCP servers and skills managed from codeg's settings, delegation, and history for the conversations you run in codeg. (#951 requested by @terranc)
- Claude Code's Remote Control works in codeg: send /remote-control in a Claude Code conversation, and a strip above it shows the session's claude.ai/code link and a QR code for picking it up in the Claude app or a browser, with a Turn off button, while the session stays connected even after you close its tab.
- Claude Code suggests your next prompt: after a turn, the empty message box shows in grey what Claude expects you to ask next, and Tab fills it in without sending.
- The side panel gains Automations, To-dos, Issues and Pull requests tabs for the project of the conversation you're in, where you can check its automations and their runs, work through its to-do board, and browse its issues and pull requests (merge requests on GitLab), with Open full page one click away.
- Agent Plan and Message navigation move from the corner of the conversation into the side panel's Session Details tab, which opens with a card showing the title, agent, status, model, Git branch and copyable session IDs, while the plan shows its progress step by step, even when folded.
- Message navigation follows the message you're reading, and clicking a message jumps there and opens it to show the sub-agents its round ran, as cards that open their conversations, and the files it changed, each with an inline or side-by-side diff.
- A sub-agent that needs you can be answered in the conversation that delegated to it: its "Waiting for task result" row opens by itself with the permission, question or plan approval inside, and its badge turns from awaiting approval to checked once you answer. (#973 reported by @lwxk)
- The model, mode and other options under the message box merge into one button that lists the current settings separated by "/", leaving out options still on their default, and clicking it opens one panel where you change each setting in place, with search for long lists such as models.
- The automation and to-do editors show an agent's settings with the same button and panel, instead of a dropdown for each option.
- A folder group in the sidebar can open or clone a folder straight into itself: hover its heading for Clone Repository and Open Folder, and the new folder lands in that group, which expands to show it.
Improved
- Updated bundled agents: Claude Code ACP 0.93.0 (Claude Code 2.1.300), OpenCode 1.19.3, Cline 3.0.73, CodeBuddy 2.168.2, Grok 1.0.62, Cursor 2026.1

This one adds Kiro CLI and ZCode as built-in agents and introduces Assistants, ready-made delegation setups with their own agent, options and standing instructions that you @-mention to hand off a task. PlantUML code blocks now draw as diagrams, a to-do that needs you can be set aside with Handle later, the side panel's tabs move to a vertical rail you can reorder, and Claude Code moves up to 0.91.0 along with seven more bundled agents.
Alongside that: an expanded Read card shows the image it read, replies in a tab left open show their full working time, stopping a Claude Code turn no longer leaves its tool cards reading Completed, Antigravity's / menu lists your skills, Windows in-app updates no longer stop on codeg-mcp.exe, delegation keeps working when codeg is started from inside an agent session, and updated components close several security holes.
New
- Amazon's Kiro CLI is now a built-in agent: codeg runs the kiro-cli you installed with Kiro's own installer, with its agent, model, thinking and effort pickers in the message box, its skills and prompts in the / menu, history import, and a settings page with Upgrade, a permission mode and an optional API key. (#851, @keh4l)
- Z.ai's ZCode is now a built-in agent: install its ACP bridge in Settings → Agents and codeg runs your signed-in official ZCode through it, with its mode, model and thinking pickers in the message box, permission requests, stdio MCP servers, delegation, and history for the conversations you run in codeg. (#941, @jackie-cqz; #685 requested by @nn200433)
- Assistants are ready-made delegation setups you can @-mention: under Settings → Collaboration → Assistants, give one a name, avatar and description, pick an agent with every one of its options and add standing instructions, and mentioning it in the message box hands it the task with exactly that setup.
- Agents can also pick an assistant on their own when they delegate, going by its description, while delegation cards, Session Details and the sidebar show which assistant ran a task, and Configuration sync carries your assistants to your other machines.
- PlantUML diagrams are drawn from code blocks marked plantuml or puml in replies, thinking and file previews, in the same card as Mermaid diagrams with zoom, fullscreen, copy and download as SVG, PNG or .puml, all rendered on your machine. (#968 requested by @f1603849451-arch)
- A to-do that needs you can be set aside with Handle later: a task waiting for review or a failed one moves to Done as it is and leaves the badge count, and from there Follow up sends it back to the agent in the same session, a failed one can still be retried, and Back to Needs you returns it.
- The side panel's tabs move to a vertical icon rail along its right edge, where you can drag them into the order you like or move a focused one with Alt+↑/↓, and every window follows the new order.
Improved
- Updated bundled agents: Claude Code ACP 0.91.0 (Claude Code 2.1.300), OpenCode 1.19.1, Cline 3.0.72, Hermes 0.22.2, CodeBuddy 2.168.0, Grok 1.0.60, Cursor 2026.10.14, Qoder 1.1.72.
- Agent defaults under Settings → Collaboration now works like the assistant editor: one agent picker marks agents that aren't installed and counts your overrides, options left on Default are dimmed, you can change the model while its other options load, and a saved value the agent no longer offers shows as unavailable instead of blank.
- Updates are checked against the version they were signed for, in the desktop app and in the server's online upgrade alike, so an older signed package can't pass for a newer one.
- The built-in model catalog is refreshed with 8,894 models from 230 providers, up from 8,831.
Fixed
- An expanded Read card for an image now shows the picture, as when Codex views one, also for images outside your workspace folders such as in /tmp, and you can open it full size, copy or download it, while a reopened Codex conversation shows the same card instead of a block of base64 text. (#977 re

This one opens PDF files right in a file tab, lets the repository panel turn issues and pull requests into to-dos on its own, and flashes the taskbar or bounces the Dock when an agent needs you. Token Usage now estimates what your tokens would cost at API list prices, and Claude Code moves up to 0.90.0 and Kimi Code to 2.2.0, along with six more bundled agents.
Alongside that: a draft can switch agents while it is still connecting, built-in browser pages on Linux stay inside the workspace window, Office previews keep their place across tab switches and wait for slow documents, message navigation reaches the whole conversation, the back gesture on phones closes what you opened instead of leaving codeg, and agents stop polling for your feedback in a loop.
New
- PDF files open in a built-in viewer in file tabs, the conversation's file viewer and canvas file cards, with zoom, fit to width, a page field, find, and a password prompt for protected files.
- The repository panel can turn issues and pull requests into to-dos on its own: under Panel settings → Auto-processing, have it scan on a schedule or on webhook events from GitHub, GitLab, Gitea or Forgejo and pick items by label and author, and it handles each item only once.
- The desktop app flashes its taskbar button or bounces its Dock icon when an agent needs you while you are in another window, on the same events as desktop notifications such as a finished turn, a permission request or a question, and Settings → General → Taskbar and Dock alerts sets each event to Off, Briefly or Until you return.
- Token Usage estimates what your tokens would cost at API list prices, for the selected range and for each model, with input, output, cache reads and cache writes priced separately from the models.dev catalog. (#503, @Adam-Dalloul)
- You can choose which tabs the @ mention panel shows and in what order under Settings → General → Mention panel, in the message box and the to-do and automation editors alike, with the Files tab always shown.
- To-dos can be searched by keyword, and #12 finds the task or issue with that number.
- The file preview header has a download button next to its ⋮ menu in the web version and in remote workspace windows, and it works for files outside your workspace folders too, such as an agent's output in /tmp. (#953, @ker2xu)
Improved
- Updated bundled agents: Claude Code ACP 0.90.0 (Claude Code 2.1.296), Kimi Code 2.2.0, OpenCode 1.19.0, Antigravity 1.4.0, Hermes 0.22.0, Cline 3.0.71, Grok 1.0.52, Qoder 1.1.71.
- Issue tasks check the report before acting: with Fix / implement or Plan first, when the report can't be reproduced, is already fixed, comes from the reporter's environment or describes intended behavior, the agent changes no code, writes no plan and gives its verdict in the first line of its reply.
- Message navigation covers the whole conversation, listing every message with its number and the files it changed, and picking one that hasn't loaded yet loads history up to it and jumps there.
- A draft conversation can switch to another agent while the one you picked is still connecting, in chat tabs and canvas draft cards alike, and the last agent you pick is the one that connects, with any startup error shown in the message box. (#963, @ker2xu)
- Sub-agent cards in Claude Code and Qoder show the sub-agent's thinking, text and tool calls in the order they happened, with thinking folded as in the main conversation, and a finished sub-agent folds all of it into one "Sub-agent activity" line with its duration. (#917, @Liang-HZ)
- When a Qoder turn breaks off while its sub-agent runs to the end, the sub-agent's card shows the result it finished with. (#917, @Liang-HZ)
- Qoder can ask you questions with its own question tool, on the same question card other agents use. (#915, @Liang-HZ)
- The model list under Agent defaults in Settings → Collaboration is grouped by provider, like the model picker in the message box, so same-named models from two providers 

This one lets agents draw charts in their reply cards, and makes Generative UI a skill you enable for each agent, now named ui-render.
Alongside that: cards in replies lose their shadow, and pressing Refresh on a Skill Packs tab no longer brings back an error or an older list.
New
- Agents can draw charts in their reply cards: ask an agent with UI Render enabled to show something as a chart, and you get area, bar, line, pie or donut, radar, radial or scatter charts in your theme's colors, with values shown on hover.
Improved
- Generative UI is now enabled per agent: Settings → Skill Packs → UI lists the UI Render skill in the same agent table as the Experts tab, so you can enable it for one agent, a few or all of them.
- Cards in replies no longer cast a shadow, so they sit flat in the message like the rest of the conversation.
Fixed
- A Skill Packs tab no longer swaps its list for an error or older data when you press Refresh while it is still loading, on the Experts, Science, Office Tools and Custom tabs alike.
- Import from agent on the Custom tab of Skill Packs no longer lists codeg's own Generative UI skill, just as it already leaves out the other skill packs.
Note
- The json-render skill is now called ui-render: start a message with /ui-render ($ui-render in Codex), and if you had Generative UI on in 0.35.0, the agents that already had json-render get ui-render in its place when you upgrade.
- The "Generative UI (json-render)" switch is gone: cards in replies always render, an agent gets UI Render in conversations started after you enable it, and the cards it already drew stay after you disable it again.

This one lets agents answer with more than text: turn on Generative UI, and a reply can show cards, tables and other interactive components right in the conversation. The chat column can be widened by dragging its edges, chat animations can be switched off, and a conversation can be maximized while files are open beside it. Claude Code moves up to 0.89.1 and Codex to 2.2.2, along with five more bundled agents.
Alongside that: the terminal panel comes back after a page reload, a workspace with many tabs opens much faster, remote workspaces keep reconnecting through outages and reopen even while their server is offline, the Open Folder dialog can create a folder, and a failed turn ends with one line saying why.
New
- Agents can reply with interactive cards and tables once you turn on "Generative UI (json-render)" under Settings → Skill Packs → UI: start a message with /json-render ($json-render in Codex) or just ask to see something as cards or a table, and the reply renders as live components while it streams.
- The chat column can be made wider or narrower by dragging its left or right edge, or under Settings → Appearance → Chat width, and double-clicking an edge puts it back to the default. (#911, @sumanx)
- Chat animations can be switched off under Settings → Appearance, so expanding, collapsing, working indicators and scrolling in the conversation happen instantly, while spinners keep turning. (#911, @sumanx)
- A conversation can be maximized while files are open beside it, from the Maximize button at the right end of its tab strip, and Restore brings the files back just as you left them.
- The Open Folder dialog can create a new folder and step into it, so a project can start from an empty folder, in the desktop app and the web version alike. (#881, @Jonathan-Asher)
- A remote workspace window can open the local workspace from Quick actions in the status bar, the macOS Window menu, or the page shown when its connection fails or expires. (#891, @ker2xu)
Improved
- Updated bundled agents: Claude Code ACP 0.89.1 (Claude Code 2.1.293), Codex ACP 2.2.2 (Codex CLI 0.160.1), Cline 3.0.70, CodeBuddy 2.164.0, Grok 1.0.51, Qoder 1.1.67, Hermes 0.21.6.
- A failed turn ends with one muted line giving the agent's own error in Codex, Claude Code, Cline and Qoder conversations, both as it happens and when you reopen the conversation.
- A workspace with many open tabs opens much faster, because a tab sitting behind another now loads its conversation the first time you switch to it instead of every tab downloading its conversation at once. (#913, @Flyneen)
- Resizing the window is smoother, as the workspace no longer runs extra full-page layouts on every step. (#898, @sumanx)
- Slash commands show the arguments they take in the / menu, the message box's Add menu and the automation editor.
- Codex answers /mcp with the live status of each MCP server, and /mcp reconnect works there too.
- Claude Code's Agent card shows the effort a sub-agent was asked to run at, next to its model.
- Claude Haiku 5.5 is now the suggested Haiku model in Claude Code's settings and in the model provider dialogs.
- On phones, the web version tints the browser bar and Android's navigation bar to match your light or dark theme, and adding it to the home screen installs codeg as an app with a full-bleed icon, even behind a sign-in proxy. (#863, @eyjohn)
Fixed
- Reloading the page or reconnecting no longer loses the terminal panel: its tabs come back with their commands still running and their output in place, and a finished command's output stays until you close its tab. (#830, @dawNotPoi; #749 reported by @kekxv)
- Claude Code no longer leaves a conversation spinning for hours after you send a message during a turn it started on its own, such as after a background task finished, and that message now gets a single reply. (#893 reported by @Luis-WowApps)
- Sending Claude Code a message while it is still writing a tool call no longer fails the turn.
- Messages you sen

This one lets you tag conversations: colored tags for every conversation or for one folder, shown after the title and ready to filter the sidebar by. The built-in browser can show a page as a tablet, a phone or any size you type, and Claude Code moves up to 0.86.0 along with six more bundled agents.
Alongside that: codeg's GitHub home is now spacering-net/codeg, a long Claude Code conversation no longer runs codeg out of memory, Claude Code answers /mcp in the conversation, and a queued message you insert as a turn ends goes out first.
New
- You can tag conversations from the conversation header or the Tags item in a sidebar conversation's right-click menu, and create, edit, reorder or delete tags in the "Manage tags" dialog.
- A tag is either global or a folder's own: global tags are offered on every conversation, chat mode included, and a folder's tags only on conversations in that folder and its worktrees.
- The sidebar can be filtered by tags, showing the conversations that have any or all of the tags you pick.
- A branch tag can show the git branch each conversation started on after its title, in a color you choose, once you turn it on in "Manage tags".
- The built-in browser can show a page as a tablet, a phone or at a width and height you type, from the device button to the right of the address bar, and each tab keeps its choice even after a restart.
Improved
- Updated bundled agents: Claude Code ACP 0.86.0 (Claude Code 2.1.287), OpenCode 1.18.35, Cline 3.0.68, CodeBuddy 2.161.4, Antigravity 1.3.0, OpenClaw 2026.9.8, Cursor 2026.10.01.
- Claude Code answers /mcp in the conversation with a list of your MCP servers, and /mcp reconnect, enable and disable work there too.
- Claude Code shows its Agent option again when you have custom Claude Code subagents, so a conversation can run as one of them from the next message on.
- Antigravity conversations show their context window usage in the status bar, which Antigravity 1.3.0 now reports.
- The built-in model catalog is refreshed with 8,414 models from 226 providers, up from 8,339 models.
Fixed
- codeg no longer runs out of memory when Claude Code writes a large batch to a conversation's transcript at once, as it can when picking a long conversation back up, which used to push codeg past 13 GB until the system killed it. (#885 reported by @andrehqh)
- Claude Code keeps your Fable pick after the update to 0.86.0, which renamed the model, instead of switching to the default model every time a conversation connects.
- A queued message you click "Insert into current turn" on goes out first if the turn ends before it gets in, with a notice saying so, instead of waiting behind the messages queued before it. (#883, @leehom0123)
- Edits you make in the message box while a message is being inserted into the current turn are kept instead of being wiped afterwards, and clicking insert on queued messages in quick succession no longer inserts two at once. (#886, @leehom0123)
- A CodeBuddy sub-agent's tool calls stay inside its Agent card while they are still being written, with CodeBuddy 2.161.4.
Note
- codeg now lives at github.com/spacering-net/codeg, and old links, update checks from installed versions and the install scripts are redirected there, so nothing needs to change.
- Docker images now go to ghcr.io/spacering-net/codeg, while ghcr.io/xintaofei/codeg stays at 0.33.0, so switch to the new address to keep updating; spacering-net/codeg on Docker Hub carries on as before.
- The browser's tablet, phone and custom views only change the page size — sites still see a desktop browser with mouse input, so one that picks its mobile version by browser type keeps showing the desktop one.
- With Antigravity 1.3.0, a "Bypass proxy for" list of just * no longer keeps all of its traffic direct — when a proxy is configured, some of its requests go through the proxy again.
Thanks to @leehom0123 for contributing to this release, and to @andrehqh for the report.

This one lets agents see and operate your desktop. With the new Computer Use preview, you share a window, a whole app or the entire screen, and agents can read it and click, type and drag in it, with Stop sharing always one click away. Codex moves up to its 2.1.1 adapter and Claude Code to 0.85.1, and seven more bundled agents get new versions.
Alongside that: the desktop app reopens the workspace windows you had open when you quit, the on-screen keyboard stays down when you switch conversations on a touch device, icons draw a little lighter, a to-do runs the agent its picker shows, and on Windows codeg-server installs into a folder of its own.
New
- Agents can see and operate the windows you share with them (preview) — switch it on under Settings → Computer Use, share a window from the status bar's Computer use panel as "Read only" or "Read and act", and agents can take screenshots, read what the window shows, and click, scroll, type and press keys in it on macOS, Windows and Linux, while the panel lists what they just did; nothing is readable until you share it. (#870)
- One click or one shortcut stops everything — "Stop sharing" in the Computer use panel or on the floating stop bar, or Ctrl+⌘+Esc (Ctrl+Alt+Esc on Windows and Linux), ends every sharing and cuts off whatever agents are doing, and a shared window no agent has used for 30 minutes (adjustable) stops being shared by itself. (#870)
- codeg's own windows and your never-share list stay off limits — password managers and system settings are on the list by default, and you can add or remove apps. (#870)
- You can share a whole app or the entire screen — an app shared as a whole includes the windows it opens later, its menus and its own shortcuts, and "Offer the entire screen" (macOS and Windows, off by default) lets agents see one picture of the screen and click anywhere on it, with codeg's windows and never-share apps painted over. (#870)
- Agents can do more than click — drag, hold keys, restore minimized or hidden windows and, for apps that take no keys in the background (such as Edge, Chrome and VS Code on Windows), bring the window to the front, which you can switch off; opening apps, moving windows and using the clipboard have their own switches, off by default, and an agent can only paste what it put on the clipboard itself. (#870)
- codeg-server can offer computer use too — start it with CODEG_COMPUTER_USE=1 and its web clients can share the server machine's windows and stop them from the same panel. (#870)
- The desktop app reopens the workspace windows you had open when you quit — the local one and each remote workspace, with the one you were using back in front, and a remote workspace that can't be reached is reported in the local window.
Improved
- Updated bundled agents: Claude Code ACP 0.85.1 (Claude Code 2.1.286), Codex ACP 2.1.1 (Codex CLI 0.159.3), OpenCode 1.18.34, Cline 3.0.67, CodeBuddy 2.161.0, Grok 1.0.46, Qoder 1.1.65, OpenClaw 2026.9.7, Cursor 2026.09.28.
- Forking a conversation no longer leaves the original session held by the agent — codeg closes it after the fork, so Codex lets go of it within about a minute and Claude Code stops keeping a process running for every fork, and the "open elsewhere" banner now says to reload in about a minute if you just forked.
- Icons draw a little lighter — those at the icon set's default weight of 2 now use 1.75, which sits better beside small text, while icons given a deliberate weight keep it. (#866, @SousekiL)
- Settings → "Browser" is now "Browser Use", next to the new "Computer Use" page, and "Collaboration" moves up to follow "General". (#870)
- On Windows, codeg-server installs into its own folder, %LOCALAPPDATA%\codeg-server, so neither install overwrites the other's files or stops the other's processes, and re-running the install script moves a server that shares the desktop app's folder out of it. (#870)
- The desktop app drops about 70 MB it never used — the standalone server binary is no longer 

This one lets Infinite Canvas, formerly Infinite Conversations, hold as many canvases as you like, each opened from a card list. Codex moves to its 2.0.1 adapter with GPT-6.1 Sol as its new default model, and an agent's settings now offer its newest release when it's ahead of the version codeg recommends.
Alongside that: a customized Codex model list picks up new official models on its own, and a to-do's base branch shows HEAD and the branch it's on.
New
- Infinite Canvas, formerly Infinite Conversations, holds as many canvases as you like — it opens on a card list that previews each canvas and lets you search, create, edit (name, description, color) or delete them, and the board you already had carries over as "Untitled canvas".
- Codex gets GPT-6.1 Sol, its new default model — it arrives with Codex CLI 0.159.2 and heads the composer's model picker.
- An agent offers its newest release when it's ahead of the one codeg recommends — its Version Status in Settings → Agents shows an amber "Upgrade to unreviewed latest X" button that asks you to confirm before installing, and now labels the recommended version "Recommended" instead of "Remote".
Improved
- Updated bundled agents: Codex ACP 2.0.1 (Codex CLI 0.159.2).
- A customized Codex model list picks up new official models when Codex is installed or upgraded, in Codex's own order, instead of only after you refresh the list and save it again.
- A to-do's Base branch reads "HEAD" beside the branch the project folder is on when you leave it at the default, instead of a grey "Current branch".
Note
- The "Adapter version" setting is gone — an agent you had set to "Latest (unreviewed)" goes back to installing and upgrading to the recommended version, so use "Upgrade to unreviewed latest" when you want the newest one.
- Codex conversations that never had a model picked now run on GPT-6.1 Sol — if the model provider you use for Codex doesn't serve it yet, pick a model in the composer or fill in the provider's Model field.

This one brings what agents build in HTML into the conversation: Codex's visualizations, Hermes' previews and the pages an agent writes open as cards right in the reply. Codex moves to its 2.0 adapter, where Read-only is read-only again next to a new Workspace access mode, and Claude Code's Sonnet is now Sonnet 5.5.
Alongside that: desktop notifications name the conversation they're about, the Recent list can show just chats or just folder conversations, the Commits tab opens on your current branch, OpenCode 2.x conversations show up in history, Grok's background workflows stay in view, and Pi can think at Max.
New
- What an agent builds in HTML now shows up in the conversation — Codex's visualizations (Claude Code's too, when it uses the same skill) and Hermes' previews appear as cards right in the reply instead of as raw markup, any other local HTML file a reply mentions gets a "Preview" row that opens into one, and a page an agent wrote starts with its scripts off until you turn them on. (#847, @SousekiL)
- The Recent list can show just chats or just folder conversations — pick All, Chat or Folders from the "⋯" button on its header; the choice is remembered, and the button stays visible while the list is narrowed. (#848, @SousekiL)
- Codex gets a "Workspace access" mode, and "Read-only" is read-only again — with Codex ACP 2.0, "Workspace access" edits files in the workspace and asks before anything beyond it, "Read-only" asks before every edit, and the default mode, where a model screens requests and asks you only about what looks unsafe, is now called "Auto review" (formerly "Approve for me").
- Claude Code runs Sonnet 5.5 — Claude Code 2.1.284 makes it the model behind "Sonnet" in the model picker.
- Claude Code's settings and model providers get a "Default Fable Model" field next to Opus, Sonnet and Haiku, which a bound provider sets or clears like the others.
- Pi can think at Max — on models that offer it: the composer lists the thinking levels the current model really has, and Pi's settings say which level Pi will run the selected model at when it doesn't offer the one you chose. (#467 reported by @Smith-106, #842, @dawNotPoi)
Improved
- Updated bundled agents: Claude Code ACP 0.84.0 (Claude Code 2.1.284), Codex ACP 2.0.0 (Codex CLI 0.158.0), OpenCode 1.18.33, Pi ACP 0.0.34, Kimi Code 2.1.1, Cline 3.0.65, CodeBuddy 2.159.0, Qoder 1.1.64, Hermes 0.21.5, OpenClaw 2026.9.6, Cursor 2026.09.26.
- Desktop notifications name the conversation they're about — its title heads the notification and its own folder leads the message, instead of whichever folder the window has open, even for a tab closed while its agent was still working; with "Hide notification contents" on, the conversation's title is left out. (#834, @Jonathan-Asher)
- The Commits tab opens on the branch you're on — it follows the current branch of the folder or worktree instead of listing every branch, and "All branches" stays selected once you choose it. (#620 requested by @Luis-WowApps, #838, @dawNotPoi)
- "Review & fix" on a pull or merge request first judges whether its approach holds up — when it doesn't, the agent commits nothing and the to-do comes back under "Needs you" with why and what to do instead, rather than with fixes piled onto a design that would have to be thrown away.
- Claude Code says when a plugin fails to load — a warning names the plugin and Claude Code's reason, instead of the plugin silently losing its commands, skills, agents or MCP servers.
- When an agent turns a message away because you need to sign in again, its notification has a Sign in button — for every agent now, not just Claude Code and Codex.
- Claude Code's context usage shows the model's real window right after a session opens or switches models, instead of reading 200K until the first reply.
- The file breadcrumb shows its folders as plain text instead of pills, spaced like the folder › title trail in the conversation header, so a file tab's header matches a conversation'

This one teaches the built-in browser two things: marking up a screenshot before it goes to the agent, and working inside remote workspace windows, where the remote host's own pages now load. Session warnings and errors now all arrive the same way — as a notification you can find again in the status bar — leaving the banner under the composer to what's still in progress.
Alongside that: the network proxy gets a bypass list and no longer breaks OpenCode and Antigravity when it runs on another machine, links in replies stop turning into "[blocked]", dragging a tab no longer empties it, and Antigravity moves up to 1.2.1, which installs on Intel Macs too.
New
- You can mark up a screenshot before sending it to a conversation — "Send to chat" in the built-in browser's address bar gains "Mark up a screenshot…" for drawing numbered boxes and arrows, and the agent also gets where each mark sits on the page, so "box 2" in your message points at something real. (#817, @xintaofei)
- Remote workspace windows get the built-in browser, and the remote host's own pages load in it — an address that lives on the remote host (a localhost:3000 an agent printed there, a container's IP) opens in a tab whose every connection leaves from the remote host through its codeg server, and a page that can't be reached says why. (#822, #824, @xintaofei)
- The network proxy gets a "Bypass proxy for" list — in Settings → System → Network Proxy, enter the domains and IP addresses to reach directly, separated by commas; localhost, 127.0.0.1 and ::1 always connect directly.
- Antigravity can be installed on Intel Macs — 1.2.1 is its first release with an Intel build.
Improved
- Updated bundled agents: CodeBuddy 2.157.0, Antigravity 1.2.1.
- Session warnings and errors now arrive as notifications instead of on the session banner — each pops up once as a toast carrying its Sign in, Retry or New session button and stays in the status bar's Alerts list, while the banner under the composer only shows what's still in progress, such as a retry with its attempt count and countdown.
- The composer's agent connection icon says why the last connection attempt failed, instead of just showing a dimmed "Disconnected".
- More agent errors read in your language — a failed mode switch or setting change, an image the agent didn't receive, or a session that couldn't be restored and started over, which now also says the agent has lost the earlier context.
- Codex's "long threads and multiple compactions" warning no longer pops up after every compaction — the compaction divider already shows it.
Fixed
- Relative file links in replies open — ./index.html no longer fails with "File does not exist", and index.html, ../docs/a.md, a.ts:12, Dockerfile or ~/notes.md no longer show as "[blocked]"; each opens from the conversation's folder, in thinking blocks too. (#815, @nnnmdzz)
- Dragging a conversation tab no longer empties its messages or closes another tab, and its context usage, session details and Diff panel keep up with the conversation after the move. (#726, @Adam-Dalloul)
- On Windows, a conversation you reopen no longer comes up with a message box you can't type in — after a restart or reload, its text area could shrink to nothing once the folder and branch row appeared beneath it, leaving only the border and buttons. (#828, @tangsenfei)
- OpenCode and Antigravity work with a proxy on another machine — their connections to their own local services now skip the proxy instead of failing with "OpenCode service failure" or "Failed to connect to WebSocket". (#804 reported by @zhou51320)
- A Codex conversation you edited or retried in the Codex desktop app shows its current history — the earlier turns followed by the new ones rather than the old version — and is listed once instead of twice. (#823 reported by @isCopyman)
- A web page opened beside a conversation over To-dos, Infinite Conversations or the Repository panel now shows, instead of leaving the side panel blank. (#824, @xintao

This one finishes 0.32.0's move onto the official ACP 2.2 runtime: an agent that crashes now says why instead of leaving behind a session that looks connected but fails every message, a permission request the agent takes back closes by itself, and no agent is left waiting on a reply that never comes.
Claude Code and Codex move up a patch as well, bringing GPT-6 Sol and GPT-6 Luna to Codex, and a server that can't update itself in place now tells you why before you try.
New
- Codex can use GPT-6 Sol and GPT-6 Luna — both arrive with Codex CLI 0.156.1 and show up in the composer's model picker.
Improved
- Updated bundled agents: Claude Code ACP 0.81.1, Codex ACP 1.13.1 (Codex CLI 0.156.1), Kimi Code 2.1.0, Grok 1.0.41, Qoder 1.1.62, Cline 3.0.64.
- A permission or question card the agent withdraws now closes by itself — for a Claude Code sub-agent you stopped or a tool that was interrupted, say — instead of lingering unanswerable and holding up the cards queued behind it.
Fixed
- An agent that crashes now says why — at launch or mid-reply, the error shows its exit status and last lines of error output instead of a bare "Incoming transport closed", and a crash mid-reply closes the connection rather than leaving a session that looks connected but fails every message.
- An agent is no longer left waiting forever on a request codeg doesn't handle — it now gets a "not supported" answer and can carry on with the turn.
- After a mid-turn message, a Claude Code turn now ends when its reply does, instead of sometimes spinning on until you stop it or send another message.
- Claude Code's Write and Edit cards no longer come up blank — when the model spells an argument differently (path for file_path, say), the card still shows the file and its content, live and in history.
- Approving a Claude Code plan moves the mode picker to the mode the session actually switched to instead of leaving it on Plan, and the approval card can now offer Bypass permissions alongside Auto.
- A Claude Code reply that merely mentions "Please run /login" no longer fails the turn as though you had been signed out.
- The "Sign in" button on a Claude Code or Codex session banner opens Settings in a window of its own, on that agent's page, instead of turning the workspace you were in into the settings page.
- A stray newline Codex sends after a reply ends or is stopped no longer leaves a "Load background activity" button behind. (#802, @Adam-Dalloul)
- A server that can't write to its own installation no longer fails its update with "close the app and try again" — Settings → System → Software Update and the status bar say which directory it can't write (or the system's reason, such as a full disk) before you try, and link the release page for a manual update. (#816 reported by @huzhiyun, #819, @dawNotPoi)
Note
- If you've customized Codex's model list, GPT-6 Sol and Luna appear only after you click the refresh button next to "Official models" in that list and save again — a customized list replaces Codex's own model table, so new official models don't join it on their own.
- Kimi Code 2.1.0 won't read, search or edit a file through a symlink that leads outside its workspace — such as a linked folder in a multi-folder workspace, or the files that come with an Experts or Office skill installed by codeg; the error names the real path, and going through that path works.
Thanks to @Adam-Dalloul and @dawNotPoi for contributing to this release, and to @huzhiyun for the report.

Claude Code and Codex move up together this time, and codeg takes up the two things both of them just learned to say: a notice now pops up as a toast instead of being written into the reply, and a context compaction shows how it really ended — and what it kept.
Underneath, codeg now talks to every agent through the official Agent Client Protocol 2.2 runtime. Server deployments can reach dev servers by hostname instead of a port range guessed in advance, and the web version gets the browser settings that still apply to it.
New
- A context compaction can show the summary it kept — when a live Claude Code or Codex session compacts, the divider gains a "Summary" toggle that opens what the model retained of the dropped history, streaming in as it is written. (#809, #814, @xintaofei)
- Server deployments can reach dev servers by hostname instead of by port — set CODEG_BRIDGE_HOST_PATTERN to auto or a template like {port}.preview.example.com, and 3000.codeg.example.com reaches port 3000 through the port codeg already listens on, with no port range to publish in advance. (#788 reported by @kekxv, #801, @xintaofei)
Improved
- Updated bundled agents: Claude Code ACP 0.81.0 (Claude Code 2.1.280), Codex ACP 1.13.0 (Codex CLI 0.155.1, which retires gpt-5.2 and gpt-5.4-mini). (#806, @xintaofei)
- Claude Code and Codex notices pop up as a toast instead of a bold line in the reply — a model fallback, a hook that blocked your prompt, a config or deprecation warning; warnings and errors also stay on the session banner. (#809, @xintaofei)
- Codex commands stream their output into the card as they run, instead of sitting on a [Terminal: …] placeholder and filling in all at once when the command ends. (#806, @xintaofei)
- Settings → Browser in the web version now holds the settings that still apply there — Local servers, Site rules and the Terminal link menu — so the notice for a server started in a terminal can finally be turned off or made to open a tab. (#788 reported by @kekxv, #798, @xintaofei)
- Writing to an agent's JSON config no longer re-sorts it — saving an MCP server, for one, leaves the rest of the file in the order you wrote it, and tool parameters likewise read in the order the agent sent them. (#814, @xintaofei)
- codeg now talks to every agent through the official Agent Client Protocol 2.2 runtime, where one message codeg fails to handle no longer drops the whole connection. (#814, @xintaofei)
Fixed
- A compaction divider says how the compaction really ended — a failed one gives its reason, an interrupted or cancelled one stops spinning, and two Codex compactions in one turn each get a divider of their own. (#806, #809, @xintaofei)
- Claude Code sub-agent cards in your history open on the sub-agent's report, not on the provenance boilerplate Claude Code 2.1.277 and later put in front of it. (#806, @xintaofei)
- The first turn after resuming a Claude Code session no longer counts the whole earlier session's tokens, so the Token Usage dashboard's per-model rows stop jumping after a resume. (#806, @xintaofei)
- Codex sub-agents no longer outlive a failed or interrupted turn — Codex 1.13.0 now closes every sub-agent the turn started, so their cards stop hanging on "running". (#806, @xintaofei)
Thanks to @kekxv for the feedback.

This one is about staying up. An agent that refuses a slash command, a grok handshake, a Cursor install on Linux, a long turn with thousands of tool calls, a crash with nothing in the log — each used to end the session you were in, and none of them does now.
Alongside that: Settings gives collaboration and the built-in browser pages of their own, the file tree can copy the file itself, and thinking blocks stop opening and closing under you mid-answer.
New
- Settings splits "Collaboration" and "Browser" out of General — delegation and in-conversation tools on one page, the built-in browser on another, unfolded.
- The file tree's right-click menu becomes a "Copy" submenu — relative path, absolute path, or the file itself onto the system clipboard (desktop only).
- Cursor's own extension calls are answered instead of refused, so a Task spawn no longer shows a red "Method not found" banner. (#783 reported by @goon-13, #785, @goon-13)
- A crash now leaves a record in the log — message, location, version and backtrace, written before the process dies. (#703, @Adam-Dalloul)
Improved
- Updated bundled agents: Grok 1.0.40, OpenCode 1.18.32, CodeBuddy 2.156.0, Cursor 2026.09.18, Qoder 1.1.60, Hermes 0.21.4.
- A thinking block stays folded until you open it, and stays open once you have.
- File and folder rows line up at every depth, with the redundant folder icon gone and the indent tightened.
- Backspace removes a reference badge in one press instead of first eating the invisible space behind it.
- The composer is usable on mobile web again — the editable area no longer collapses, and a tap anywhere in it opens the keyboard. (#746 reported by @evepupil, #779, @evepupil)
- File attachments survive a reload on custom ACP agents and grok, and an attachment-only chat is named after the file. (#780, @tangsenfei)
- A reply's footer keeps filling in for agents that flush late, so deepseek replies get their model, tokens and time — and "fork from here" stays available.
- The default-terminal picker says what your choice actually resolves to, and badges a shell that is not installed.
Fixed
- An agent rejecting a prompt no longer tears the session down — Qwen Code refusing /mcp left the composer greyed out until a full respawn. (#797 reported by @linshaobao)
- Grok 1.0.40 connects again — its new setup frames carry no session id yet, which read as a protocol error. (#794 reported by @qingyueyin)
- Cursor works on Linux, and an expired login now says so instead of reading green while every session fails.
- Everything after a Claude Code /clear stays visible on reopen. (#766 reported by @andrehqh, #778, @Frank-zhu0404)
- A long turn no longer freezes the session — its tool-call history was resent whole on every attach (26.7 MB measured). (#380 reported by @ashlovepink, #712, @Adam-Dalloul)
- An agent reply no longer renders blank, which the sub-agent dialog hit on an ordinary path. (#705, @Adam-Dalloul)
- A message sent mid-turn no longer duplicates the first half of the reply. (#708, @Adam-Dalloul)
- Built-in browser: a Google sign-in popup leaves no empty tab behind, a page redirecting mid-load is no longer covered by an error page, and the first blank tab stops spinning.
- A deleted or manually closed Office preview no longer reopens itself. (#795, @Adam-Dalloul)
- Codex sessions are titled after your message, not after the AGENTS.md fragment injected ahead of it. (#789 reported by @zhoujh78, #790, @dawNotPoi)
- The scientific research pack installs for Antigravity, Grok, Cursor, DeepSeek and Qoder, whose toggles used to flip themselves back off. (#718 reported by @sunnyhmz7010, #793, @dawNotPoi)
- Launching at Windows login works within the first half-minute after boot. (#703, @Adam-Dalloul)
- A Telegram group message can no longer crash the polling loop. (#703, @Adam-Dalloul)
Thanks to @Adam-Dalloul, @goon-13, @evepupil, @tangsenfei, @dawNotPoi and @Frank-zhu0404 for contributing to this release, and to @linshaobao, @qingyueyin, @andrehqh, @a

This patch smooths out the built-in browser introduced in 0.31.0: agent-written JavaScript can run without repeatedly interrupting you after you explicitly enable it, browser content now reaches the edges of its pane, blank tabs match your theme, and Windows workspaces keep the correct size after a browser tab is closed.
Improved
- Agent-written JavaScript no longer asks for approval before every snippet by default — it still runs only after you enable the browser tools and code-execution switch and share the page with “Read and act” access, every run remains visible in the tab’s activity, and you can choose “Ask me every time” under Settings → Built-in browser → Running code on a page.
Fixed
- Browser pages and HTML previews now extend cleanly to the edges of their pane — the permanent thin frame around them and the one-pixel gap left behind when the terminal was closed are gone.
- Blank browser tabs now follow the app theme — dark mode no longer opens onto a bright white page, theme changes apply immediately, and in-app notifications remain visible above the blank tab.
- Windows workspaces continue to resize correctly after a browser tab is closed — closing a tab while the window is maximized and then restoring it no longer leaves the app drawn at its maximized size with the title bar and window controls pushed beyond the right edge.

codeg has a browser now. A link in a conversation, a tool result or the terminal opens as a tab beside your files — with profiles, downloads, find in page and a locked-down preview for local HTML — instead of throwing you into another app.
It is also a surface you and an agent can share: hand a page, an element, a screenshot or a console error straight into the conversation, or let an agent read and drive a page you have shared, with everything it does written on the tab where you can watch it.
Alongside it: the blocking question card folds away to its header, qoder sessions finally report what they cost, and a cancelled task put back on the board starts over cleanly.
Built-in browser
- Links open in a tab beside your files instead of leaving the app — from a conversation, a tool result, the terminal, the editor or a notification, on macOS, Windows and Linux; ⌘/Ctrl-click uses your system browser just that once, and Settings → Built-in browser sets the destination per source, per site, or turns the whole thing off. (#723)
- A tab is a real browser — back, forward, reload and the address bar in one row, find in page with ⌘F, downloads into your downloads folder, pop-ups that say why they were blocked and offer to open anyway, and an error page that names what went wrong and offers Retry.
- Tabs come back after a restart, and can hand their memory back after half an hour in the background if you would rather have the memory than the page's state.
- Profiles keep two sign-ins apart — separate cookies, storage and logins, a choice of which profile new tabs use, the same page re-openable in another, and browsing data clearable one profile at a time; signing in with Google from a tab works too, which embedded browsers are normally refused.
- A server you start in a terminal comes to you — codeg notices the address it prints and offers it in a toast (or opens a background tab, or stays quiet — your choice), only ever for addresses on this machine, and the “+” at the end of the tab strip lists the local servers running right now.
- HTML files preview in a locked-down document view — no scripts until you press Enable, and even then only over files in the document's own folder with no way to reach the network; the old inline preview stays one switch away.
- Hand what you are looking at to the conversation — the page, an element you pick on it, a screenshot, or the errors its console printed, each landing in the composer as a badge you can still take back before sending.
- Let an agent work on the page in front of you — with the "Read and drive the built-in browser" tool group switched on (it is off for every agent until you do), an agent can list your open tabs, read a shared page's structure, click, type, press keys, choose options, screenshot it, read its console, and open, navigate or close tabs.
- And see exactly what it did — each tab's address bar says what it is sharing and takes it back in one press, every action an agent takes is written on the tab itself with refusals included, sharing ends the moment the page leaves that site, and the standing level for new pages (read and act, read only, or nothing) is yours to set.
- An agent running its own JavaScript is a separate switch, off by default — each snippet is shown to you for approval before it runs, and approving one never approves the next.
- Platform notes — on Linux a tab is a window of its own rather than a pane inside the workspace, and the locked-down HTML document view is macOS and Windows only. Favicons, download progress and background preloading are not in this first release.
New
- Collapse the blocking question card to its header — one click gives the message list its space back while the question stays pending, and the card re-opens itself if the agent replaces the question set or your answer comes back failed. (#622, #623, @yunsheng111)
- The composer's token popover shows your cache hit rate — one row under Total, to one decimal, with the Token Usage dashboard a

Your configuration can now travel between machines — model providers, agent settings, custom agents, quick messages and task templates, as a file or through your own WebDAV server, optionally encrypted.
A fast reply no longer locks up the interface: the streaming window now scales with the answer it re-renders, and a two-minute reply at full speed costs about a tenth of the rendering work it used to.
The composer learns two shortcuts — recall a sent prompt with the arrow keys, and click a queued message into the turn that is already running — while Gemini CLI, OpenCode, Cline and Kimi all get a deep adaptation pass.
New
- Configuration sync between machines — a configuration-only snapshot of model providers (with their keys), agent settings, custom agents, quick messages, task templates and app preferences, carried by an exported file or by a WebDAV server you already own, on the desktop and in server mode alike; conversations and uploads stay out of it, applying a snapshot never deletes what the target machine has, and automatic sync only ever uploads. (#747, @galact-byte)
- The uploaded snapshot can be encrypted with a passphrase (AES-256-GCM), and both the WebDAV password and the passphrase live in the system keyring instead of the database.
- Every import and restore leaves a way back — the last ten configurations this machine had are kept, reachable from the file-actions row, and undoing one is itself undoable.
- Recall a sent prompt with ArrowUp / ArrowDown in the composer, the way a shell history works. (#744 requested by @evepupil, #756, @dawNotPoi)
- Click a queued message to insert it into the running turn — on any session with a live-feedback channel, instead of waiting for the turn to end. (#751, @leehom0123)
- A work task can choose the branch it is created for — the worktree starts at that branch's tip, the merge lands back on it, and a delivery opens its pull request against it, so a task no longer depends on what the project folder happens to be checked out on. (#760 requested by @wht300)
- OpenCode has a "Behavior & compaction" settings card — default agent, sharing, auto-update, filesystem snapshots, sub-agent depth, auto-compaction and tool-output limits, written to opencode.json, with any field left unset still following OpenCode's own defaults.
- Cline's provider settings now split the two ways cline actually authenticates — Cline Usage-Billing, ClinePass and ChatGPT get the cline auth <id> command to run instead of a key field, bring-your-own providers keep their key fields, and anything configured outside codeg keeps its own row.
- Gemini sub-agent transcripts now show up — a sub-agent's own conversation is discovered alongside the session that launched it.
- OpenCode's lsp is now a permission of its own, so you can decide it separately from the rest.
Improved
- Updated bundled agents: Claude Code ACP 0.78.0 (Claude Agent SDK 0.3.270), Codex ACP 1.12.0 (Codex CLI 0.154.0), Gemini CLI 0.60.0, Kimi Code 2.0.0, Grok 1.0.34, Qoder 1.1.54.
- A reply streaming at full speed no longer freezes the interface — the flush window now grows with the answer being re-rendered and is timed per conversation, taking a two-minute reply at 300 tokens/s from 519M re-rendered characters down to 60M while one busy tab stops holding back the others. (#589 reported by @Lparksi, #716, @Adam-Dalloul)
- Settings → System is now one "Data & sync" card — configuration sync, backup and restore as three tabs, with the pre-restore snapshots sitting under the restore flow they undo instead of in a card of their own.
- Kimi finds skills it was already loading — codeg now reads all four of Kimi's skill roots, including the shared store pi and cline install into, and writes project skills at the repository root where Kimi actually looks.
- A kimi-k3 session's context gauge is correct — history was measured against a 256K window on a model that has 1M.
- Opening a historical conversation now shows the session coming up — the composer names t

codeg:// is now a real system link — an outside tool can hand the desktop app a conversation and land you on that tab, cold start included.
The close button stops guessing (ask, minimize to tray, or exit — your call), and an agent's temp files no longer pile up: every launch gets its own scratch directory that's cleaned when it ends.
Cline connects with your own provider again, CodeBuddy's background workers report into the card that launched them, and eleven bundled agents move up.
New
- codeg://session/<id> opens that conversation from outside the app — on macOS, Windows and Linux, whether codeg is already running or starts cold. (#725, @changexbc)
- The close button is now a preference — ask every time, minimize to tray, or exit; the prompt has a "remember my choice" box, and Settings → General keeps it editable afterwards. (#633 requested by @galact-byte, #741 @galact-byte)
- Antigravity's settings panel can reclaim leftover temp space — scan the system temp folder for directories older versions stranded, review each path, and delete only what you confirm; anything still in use is skipped. (#735 reported by @chenjing-x)
Improved
- Updated bundled agents: Claude Code ACP 0.77.0 (Claude Code 2.1.270), Codex ACP 1.11.0, Kimi Code 0.43.1, CodeBuddy 2.151.0, Grok 1.0.30, Qoder 1.1.53, Cursor 2026.09.10, Hermes 0.21.3, OpenCode 1.18.31, OpenClaw 2026.9.4, Cline 3.0.62.
- The model and reasoning-effort selectors now mark the agent's own recommendation — Codex and Claude Code each name their default, badged "recommended" in all ten languages.
- Claude Code's selectors drop the ambiguous "default" row and no longer replay a saved pick the agent has stopped offering, so each turn records the model it actually ran on.
- A permission request Claude marks as not-to-be-approved-by-a-stray-keystroke now highlights Reject instead of Allow.
- An agent no longer leaks its temp directory on every connect — each launch unpacks into its own scratch directory that is removed when the agent exits, and anything a crash stranded is swept at the next start. (#735 reported by @chenjing-x)
- Downloaded agent binaries now survive a restart or a system cleanup tool — the managed binary cache moved out of the system cache folder, and existing installs are migrated in place. (#736 reported by @tyonia)
- Managed agents are found more reliably — codeg puts its own install directories first on PATH, recognizes ~/.opencode/bin, and reads both OpenCode plugin layouts (1.18.30 moved the package directory). (#736 reported by @tyonia)
- DeepSeek's session modes, options and approval buttons now follow your language — deepseek-acp advertises all of them in Simplified Chinese only, so codeg re-labels them at render time; your own model names are left untouched. (#733 reported by @Mistic92)
- A "failed to parse JSON" error now names where it came from and no longer quotes the raw payload back at you, so a credential can't ride out inside an error message.
Fixed
- Cline sessions connect with your own provider again — they failed with "Authentication required" whatever was configured, and the settings panel now writes where Cline 3.x actually reads, so a save after the first is no longer silently dropped.
- Cline's "Auto-approve tools" chip actually toggles now. (#709 reported by @udagos)
- A CodeBuddy background worker reports into the card that launched it — no more second bubble full of raw <task-notification> XML, and the card carries the worker's status, summary and its own tool calls.
- A turn an agent ran on its own no longer leaves the session looking frozen — a pill above the thread pulls it in, instead of you having to reopen the conversation.
- DeepSeek sessions written by deepseek-acp 0.9.0 open again — they rendered empty, or were silently cut off at a migration point, because the session log filename changed in that version. (#734 reported by @fanliang123)
- An MCP call with a query argument keeps its own name instead of being shown as "WebSearch" — cod

pi gets a deep pass — reasoning, images, real diffs and rewound branches now come through correctly, and a new session no longer opens with a page of startup noise.
Local screenshots render inline in replies, imported worktrees land under their repo, and six bundled agents move up.
New
- Local images in a reply now render inline instead of a blocked placeholder, and click to open in the image viewer. (#691, @Adam-Dalloul)
- pi is now a local MCP source — servers in ~/.pi/agent/mcp.json show up in MCP settings and can be edited there. (#653 requested by @galact-byte, #673 @dawNotPoi)
Improved
- Updated bundled agents: DeepSeek 0.9.0, Kimi Code 0.42.0, OpenCode 1.18.30, CodeBuddy 2.149.0, Grok 1.0.25, Qoder 1.1.49.
- DeepSeek's model panel follows 0.9.0 — the built-in list is now deepseek-flash and deepseek-v4-pro, image detail is replaced by a pixel budget, and a list written by an older codeg says so and is repaired by saving it again.
- pi's settings panel now warns that a custom config or session directory only applies to pi itself — sessions written there won't appear in codeg's history.
Fixed
- pi's reasoning is no longer dropped — thinking blocks were read from the wrong field and vanished entirely.
- A rewound pi session now shows only the branch you're on, not the abandoned ones still sitting in the same file.
- Opening a pi session no longer starts with pi's startup banner as the assistant's first message.
- More of a pi transcript survives: images, an edit's real diff, cancelled commands marked failed, an error message on a failed turn, and context-compaction cards.
- A Codex code-mode MCP call keeps its identity when you reopen the session — real tool name, arguments, result and error state, instead of a generic script card. (#702, @asteroida123)
- An imported worktree session now lands under its repo, so it joins the repo's worktree group and gets its branch label. (#552 reported by @Luis-WowApps, #707 @Adam-Dalloul)
- An agent process is no longer left running when a connection ends at the wrong moment. (#701, @asteroida123)
Thanks to @Adam-Dalloul, @asteroida123 and @dawNotPoi for contributing to this release, and to @Luis-WowApps and @galact-byte for the reports.

Codex's team-of-agents capsules stop being empty frames — a sub-agent's outcome, its closing report and its own session are all on the card now.
DeepSeek's model list becomes an editable panel, six bundled agents move up, and Antigravity's sign-in link comes back in seconds instead of never.
New
- DeepSeek's model list is now editable in Agent Settings — one row per model (id, display name, context window) with description, output cap and image limits behind "More", a "Use agent defaults" button to drop the stored list, and a notice when the harness is refusing a list you wrote by hand and quietly running on its built-in models instead.
- A Codex team-of-agents sub-agent now shows how it ended, what it reported and where it worked — the capsule keeps its id after reopening, wears the child's outcome as a chip, folds the child's closing report into the card, and offers a "View sub-agent session" button that opens the child's own transcript.
Improved
- list_agents now renders as a roster card instead of a wall of escaped JSON, carrying each finished child's full report — often the only readable copy of that text.
- Updated bundled agents: Gemini CLI 0.59.0, OpenClaw 2026.9.3, Hermes 0.21.1, CodeBuddy 2.147.0, Grok 1.0.24, Qoder 1.1.47.
- OpenClaw now asks for Node 24.16.0 or newer, since 2026.9.3 retires the Node 22 lane — the check happens before the install rather than letting it pass and the launch fail.
Fixed
- Antigravity's "Get sign-in link" returns a link again — 1.1.1 prints it to stderr, which codeg wasn't reading, so the panel spun for ninety seconds and then reported that no link had appeared; it now arrives within seconds.

Files and terminals join the infinite canvas, and a right click in the composer now acts on the link or path under the pointer.
Agent commands stop forcing color into everything they run, and merging a task no longer waits out a context compaction.
New
- Files and terminals can now be placed on the infinite canvas beside conversations and notes, and a terminal card keeps its shell running while you are away from the board.
- Right-clicking a link, an email address or a path in the composer now selects it, with a row to open the link, write to the address or open the file. (#672, @Adam-Dalloul)
- A file badge in the transcript can now download the file, for web mode and desktop windows bound to a remote workspace.
- Antigravity gets a Sign out button in its settings panel, so the Google account it silently reuses can be swapped for another. (#666, reported by @dale0525)
Improved
- The codeg-mcp popover's tool rows now follow the app theme instead of each wearing a fixed color of its own.
Fixed
- Colored command output is now opt-in — a "Colorize command output" switch in Settings → General, off by default, because forcing it also broke parsed output such as gh --json piped into jq. (#664, @chadnorvell)
- A Cline task whose message quotes </environment_details> no longer hangs codeg when opened, and a heading right after a task-progress block is no longer deleted with it. (#670, @MaxFreedomPollard)
- Ctrl/Cmd+Shift+T now reopens a closed tab where it was, not at the end of the strip. (#667, @Adam-Dalloul)
- Merging a finished task no longer waits on a context compaction — the dialog closes as soon as the round starts, the card says it is compacting, and the session viewer shows the live transcript instead of the previous round's.
- A /compact in a Claude session now looks the same live and after reopening — your command, the divider and the continuation summary.
Thanks to @Adam-Dalloul, @chadnorvell and @MaxFreedomPollard for contributing to this release, and to @dale0525 for the report.

Diffs now show images as pictures, the repository panel speaks Gitea, and workspace backgrounds get a wallpaper market.
Desktop notifications become a real setting, and backup & restore stops losing custom-agent conversations.
New
- A wallpaper market for the workspace background — browse the public Wallhaven SFW gallery from Settings → Appearance, and one click downloads and applies a wallpaper. (#626, @user-A100)
- Image files now diff as before/after pictures, with a swipe slider, in the workspace diff tab, the commit dialog, the push workspace and the unstash dialog.
- The repository panel now works against Gitea and Forgejo, with a Gitea accounts section under Settings → Version Control.
- Desktop notifications are now configurable in Settings → General: a master switch, when to notify, whether to hide contents, and one switch per event.
- A deleted conversation can be restored from the session picker — turn on "Include deleted", and importing one brings the original conversation back with its history intact.
- The status bar reports the codeg-mcp service and lets you switch its tool groups — delegation, live feedback, ask-a-question, session lookup, automations and work tasks — right in the popover.
- Claude now marks where its context was compacted, live and after reopening, saying whether it was triggered manually or automatically (claude-agent-acp 0.75.1).
- Every restore is snapshotted first and can be rolled back, with real progress, a cancel button and a result panel.
Improved
- Updated bundled agents: CodeBuddy 2.146.0, Qoder 1.1.45, Gemini CLI 0.58.0, OpenClaw 2026.9.2, and the Claude adapter (claude-agent-acp 0.75.1).
- Settings → General now opens with the notification and sound sections folded, instead of filling the page before anything below them.
- Refreshed model catalogs — Codex gains gpt-6-astra, OpenCode grows to 213 providers and 7,562 models, and gpt-6 sessions finally show a context meter.
Fixed
- A backup no longer loses custom ACP agents' conversations — their message text was packed by neither the backup nor the restore path, so a round trip left every conversation in place and empty.
- OpenCode, Hermes, Cursor and Antigravity session databases now survive a backup round trip whole, instead of losing their newest messages and being restored beside a stale write-ahead log.
- Forking a Claude conversation from a branch an earlier fork left behind now lands on the reply you picked, instead of quietly falling back to the end of the conversation (claude-agent-acp 0.75.1).
- A failed task push-back now reports git's own reason, instead of always blaming the fork author for not allowing maintainer edits.
Thanks to @user-A100 for contributing to this release.

Talking to an agent while it is still replying now works on every session that has a delivery channel — with attachments, shown as a real message in the transcript, and recoverable if the agent finished without reading it.
Mobile gets a terminal key bar and a ⋯ menu on every file-tree row.
New
- Send a message into a running turn on any session with a delivery channel — Codex, Grok, Gemini and OpenCode sessions get the mid-turn send too, worded honestly per channel: Claude inserts into the current turn, the others leave a note the agent picks up at its next check, shown as waiting until it actually does. (#637, @Adam-Dalloul)
- A message sent mid-turn now reads as a message — it lands in the transcript as its own turn, and the reply to it starts a new turn instead of running into the previous answer mid-paragraph. (#636, @Adam-Dalloul)
- Mid-turn sends carry images and file references, instead of going dark the moment you stage an attachment. An image shows on the message right away in the running turn rather than only after reopening the conversation, and if the turn ends first the whole draft — attachment included — is queued for the next one rather than dropped. (#640, @Adam-Dalloul)
- A note the agent finished without reading is no longer lost — the list stays up after the turn, says the agent never got to it, and offers to send it as an ordinary message or dismiss it.
- "Adapter version" is now a per-agent setting — follow the newest adapter release instead of waiting for the reviewed pin, which is what makes a brand-new model reachable the day it ships. Pinned stays the default and recommended; the newest release is unreviewed and can break the agent, and an install that can't fetch it falls back to the pinned version rather than failing. (#635, @Adam-Dalloul)
- A virtual key bar for the terminal on mobile — ESC, TAB, CTRL, ALT, arrows, Home/End and PgUp/PgDn, with CTRL/ALT latches that wrap the next key you type, lifted clear of the soft keyboard and collapsible from the tab bar. (#647, @mrytsr)
- Every file-tree row gets a ⋯ button that opens the same menu right-click does, so touch devices have a way in now that long-press is reserved for drag. (#648, @mrytsr)
- Transcript file links open beside the conversation when the workspace column is off screen — on the task board and the infinite canvas, a file badge, a markdown link or "view diff" now opens a read-only viewer next to the transcript, with "Open in workspace" one click away, instead of doing nothing visible.
- Codex reports its background terminals (codex-acp 1.10.0) — a long-running command Codex pushes to the background is badged "Background" and joins the live task strip with its Stop button and output link, the way Claude's already did.
- "Disable hardware acceleration" now works on Linux, not just Windows.
Improved
- Updated bundled agents: Kimi Code 0.41.0, CodeBuddy 2.144.0, Qoder 1.1.44, Cursor 2026.09.02, Antigravity 1.1.1, OpenCode 1.18.29, plus the Claude and Codex adapters (claude-agent-acp 0.74.0, codex-acp 1.10.0).
- Antigravity now takes a custom version in Agent Settings, like the other downloaded agents — and a version that was never published now fails the download instead of quietly filing the pinned build under the number you asked for.
- Signing out of Claude mid-session now ends only that turn — the session stays open and asks to be signed in again, instead of the connection dropping and the conversation flipping to cancelled. A delegated sub-agent in the same state reports "needs sign-in" rather than an unrecognized failure.
Fixed
- A Codex search that finds nothing now says "No matches" instead of a red failed tool call. (#651, reported by @rubybrowncoat; fixed by @dawNotPoi in #652)
- Cleaning up a task with no changes now removes its worktree completely, instead of half-removing it and then reporting it still holds uncommitted files so the retry never finished. (#642, reported by @rubybrowncoat; fixed by @dawNotPoi in #650)
- Worktree 

Conversation forks are now more reliable, staying on the reply and session you chose while carrying their settings into the next prompt.
Pi questions, MCP settings, and Windows network-share launches are more dependable too.
New
- DeepSeek can now fork a conversation at the exact reply you picked (deepseek-acp 0.8.0), instead of falling back to the end of the session.
Improved
- “Fork from here” is now the single, clearer way to branch a conversation — the redundant composer shortcut is gone, and the reply action stays visible with an explanation while a response is still running.
Fixed
- Conversation forks now reliably open on the intended new session and are ready for the very next prompt, preserving the selected turn plus supported model, effort, and mode settings without jumping back, hanging, or losing a fast follow-up.
- Pi’s multiple-choice prompts now use the same interactive question card as other agents, and the choice you make remains visible in the transcript. (#644)
- One unreadable agent config no longer takes down the MCP settings page — healthy configs still load, empty files are handled normally, and codeg identifies the problem and pauses writes until every existing setting can be preserved safely. (#646, @dawNotPoi)
- Windows agents launched from UNC workspaces now start in the correct project directory, including .cmd and .bat launchers used by npm-installed agents in WSL or network-share projects. (#638, @damiandelmas)
Thanks to @dawNotPoi and @damiandelmas for contributing to this release.

Fork a conversation at any reply, not just at the end — and Codex can fork at all now.
Claude's background work now shows up as a live strip you can watch, read, and stop.
New
- "Fork from here" on any finished reply — the new session keeps everything through that reply, the original is untouched, live or reopened alike.
- Codex sessions can be forked (codex-acp 1.8.0), and a fork opens with the full history it inherited.
- Claude's background tasks get a live strip at the top of the conversation — last tool, token cost, a Stop button, and an Output link that opens the log in a file tab, in the browser too.
- Background work codeg couldn't follow now settles itself: workflow and monitor tasks show up, finished ones leave the strip, and running ones keep their agent from being shut down as idle.
Improved
- Updated bundled agents: Cline 3.0.61, OpenCode 1.18.26, Kimi Code 0.40.1, Qoder 1.1.41, Cursor 2026.08.31.
- Claude Code's approval buttons now say what you're granting and for how long — this session versus always in this project (claude-agent-acp 0.73.0).
- Claude Code's session title now comes straight from the agent, so the sidebar renames the moment it's named instead of at the end of the turn.
Fixed
- A Codex Plan-mode plan shows once, as a card — no longer lost on reopen, no invented "Implement the approved plan." message, no duplicate while watching live.
- A reply that merely mentions <proposed_plan> is no longer cut off and swallowed into a "Proposed plan" card, whichever agent wrote it.
- One stretch of Codex thinking reopens as one thinking card, not a column of up to 28.
- An expanded canvas card comes back intact from another page — full transcript, still live if its agent is working, and no longer reclaimed a minute later.
- A new canvas card keeps the first message you send it, and you can now place the cursor and select text inside it.
- A delegated sub-task's working directory no longer turns into a sidebar project, reopening folders you'd closed and reshuffling their order.

Your conversations get an infinite board — lay them out in space and run several agents side by side on it.
Multi-agent collaboration gets sturdier too: an @-mentioned agent is delegated through codeg as a session you can open, and a sub-task cut off by a cancel or a crash can be picked up where it stopped.
New
- Infinite Conversations: a board that lays your workspace out in space — folder and agent regions, curated collections, sticky notes, and pinned cards that open into real working conversations, so several agents run side by side on one board instead of one tab at a time.
- Curate the board by dragging: pull a card out to pin it, drop it into a region to collect it, colour anything, auto-arrange, export a PNG, or fill the whole board from your open workspace in one click.
- Folders group in the sidebar — named, optionally coloured bands that share one order with loose folders so the two interleave, each heading carrying a running-session badge summed across everything inside it.
- The branch popup under the composer lists the branches your other worktrees are on, main working tree first and the rest folded under shared prefixes like task/.
- Cursor's model picker is families plus knobs instead of two hundred ids, with Thinking, Effort and Fast beside the family name, each shown only where that family varies on it. (#607, @ZY-Dragon03)
- A delegated sub-task cut off by a cancel or a crash resumes where it stopped, under its original id and in the same conversation, with status and cancel still working on the id you already hold.
- An agent you @-mention in a prompt is delegated through codeg, so the child becomes a session you can open instead of one the agent quietly spawns outside the app.
- Tasks can compact their context before a resumed round, at a threshold you set, with the before/after occupancy recorded on the task timeline.
- Cancelling a task can take its worktree with it, and skipping that checkbox no longer strands the checkout — the drawer offers the removal on its own afterwards.
- A repository on a host that is neither GitHub nor GitLab says so up front, instead of spending the request and reporting whatever the wrong API answered.
- Ctrl+Shift+C copies in the terminal, Git Bash on Windows included, and the terminal keeps focus afterwards. (#536, @OpenX123)
Improved
- Updated bundled agents: OpenClaw 2026.8.1, Hermes 0.21.0, CodeBuddy 2.143.0, Grok 1.0.13, Qoder 1.1.40.
- The chips above the composer explain themselves on hover with a themed hint that names what the chip is and carries the setting's own blurb.
- Issues and pull requests with hundreds of comments open quickly, rendering what's on screen and loading the rest as you scroll.
- The Reasoning Effort dropdowns are gone from the Claude Code and Codex settings pages, since the composer's per-session picker is what actually decides how hard a turn thinks — and the Codex one additionally wrote a model_reasoning_effort = "high" you never asked for whenever you saved any other setting.
- With "Show worktree branches" on, a repository's own row names its branch — main [ root ], lined up with its worktree siblings, instead of a bare "root".
- The status-bar launcher drops its "Sessions" group, whose two rows act on a single folder the launcher doesn't have and already live in the folder's context menu and on the Folders header.
- The repository panel remembers the Issues / Pull requests switcher across sessions, and drops its Beta mark.
Fixed
- A self-hosted GitLab no longer fails with 410 — an instance whose hostname doesn't announce it was treated as GitHub Enterprise, so codeg asks the instance what it is now. (#611)
- An issue you file from the repository panel appears in the list right away, without a manual refresh.
- The macOS icon is no longer a size bigger than everything else in the Dock. (#610)
- Finished replies stop growing stray * and _ after patterns like foo/* and _meta. (#487, @Adam-Dalloul)
- A message that @-mentions an agent reopens 

Read a pull request and merge it without opening a browser.
Every assistant reply now sits under a header you can fold away.
New
- A pull request opens as three tabs, and there's a Merge button. Conversation, Checks and Files changed; click a file to see its diff; the merge menu lists only what the repository allows — merge commit, squash, or rebase.
- Write a comment, close or reopen an issue, and file a new one — all in the repository panel. These used to mean going to GitHub or GitLab in a browser.
- Every reply has a header: "Working…" while the agent writes, "Worked for 1m 9s" when it's done. Click it to fold the whole reply away; sending your next message folds the one above it. A reply that stops on a tool call keeps everything on screen.
- Hover a session and a card floats out to the right with its full title, agent and model, folder, branch and path. Sessions started inside codeg now show a model too — before, only imported ones did.
- Ctrl/Cmd+1 through 9 jump to a tab, and Ctrl/Cmd+Shift+T reopens the one you just closed. (#502, @Adam-Dalloul)
- VS Code joins the "Open in" menu, for folders, directories and files. (#591, @markos-ttl)
- Antigravity can sign in on a machine with no browser. codeg gives you a link, you sign in to Google on your phone or another computer, then paste the address it lands on back into codeg.
- A conversation whose session was archived says so, and hands you the command to get it back. One click copies codex unarchive <id>; before, it just said "Failed to load session, starting new". (#598, @asteroida123)
- Codex now says why it needs your approval, instead of a fixed title like "Edit files". Comes with the codex-acp 1.7.0 upgrade.
Improved
- Updated bundled agents: Gemini CLI 0.57.0, OpenCode 1.18.25, Hermes 0.20.6, CodeBuddy 2.141.0, Kimi Code 0.39.1, Qoder 1.1.33.
- Kimi Code takes MCP servers again, which is what unblocked the jump from 0.36.1 to 0.39.1; its login hint now includes the --region global a kimi.ai account needs.
- Saving MCP settings leaves fields codeg has no UI for exactly as they were, instead of dropping them.
- The pull request detail panel lines its description, comments and composer up on one left edge, keeps the author's avatar next to you down a long comment, and is wider.
Fixed
- The interface grows when you zoom the window. At 150% the text got bigger, but column widths, small labels and popovers stayed at their 100% size.
- Chinese, Japanese and Korean text isn't clipped at high zoom in the sidebar, conversation titles and dropdowns.
- One stretch of thinking stays one card while sub-agents are streaming beside it, instead of being cut into several. (#494, #586, @dawNotPoi)
- A reply you're watching in a sub-agent transcript stops being stuck on "responding". Its timing stats and artifacts card show up.
- An image card keeps its filename or page name after a reload, rather than reverting to "Image generation". (#504, @Adam-Dalloul)
- An OpenCode tool card shows the result, not a blob of raw JSON. Watching live, a question you'd already answered read as "no selection".
- Sessions Codex delegates to sub-agents no longer show up as separate sessions in your history. (#593, @TroyMitchell911)
- When a turn produces nothing, nothing comes back — not the previous turn's answer.
Thanks to @Adam-Dalloul, @asteroida123, @dawNotPoi, @markos-ttl and @TroyMitchell911 for contributing to this release.

Diffs can be read side by side.
Mermaid diagrams get a viewer of their own.
New
- Diffs open side by side, with a toggle on every preview and each side scrolling on its own. (#537, #570, @asteroida123)
- Mermaid diagrams get their own viewer — sharp at any zoom, draggable, fullscreen, and saveable as SVG, PNG or .mmd.
- Select a passage in a message and ask about it in a new conversation, pinned to the agent that wrote it.
- The repository panel shows an item's comments, not just its description.
- Remote workspace connections can carry custom HTTP headers, for a codeg-server behind a proxy that demands one. (#577, @codebutler)
- The window zooms to 300%, and Ctrl/Cmd + and − step it. (#484, #569, @Adam-Dalloul)
- Custom Codex models work on OpenAI-compatible gateways, through a template that sends plain Responses requests.
- Codex can ask you questions in Default mode, via a new switch in its settings panel.
- A delivered task can remove its worktree on the spot, with the same checkbox merge and complete already offer.
Improved
- Updated bundled agents: OpenCode 1.18.23, Cline 3.0.60, CodeBuddy 2.139.0, DeepSeek Harness 0.7.0, Qoder 1.1.31.
- DeepSeek Harness resumes by resolved path, so /var against /private/var or a Windows short name no longer fails.
- The Recent list folds back to its first page instead of staying long for the rest of the session.
- An issue task has to confirm the problem before fixing it, and the separate "investigate only" mode retires.
- Copy image works on pictures staged in the composer. (#556, @Adam-Dalloul)
Fixed
- A task counts its own work, not the pull request it started from or the new files it never committed.
- A task waiting on your answer stops reading as "running". (#567, @lizzjin)
- A session renamed in Claude Code keeps its name. (#566, @Adam-Dalloul)
- Closing settings brings back a workspace hidden to the tray, instead of leaving nothing on screen. (#565, @Adam-Dalloul)
- The add-to-chat pill stays inside the editor on selections near the top of a file. (#576, @dawNotPoi)
- Windows paths stop mixing separators, like the clone dialog's C:\work/codeg.
- Connection secrets stay out of the logs, where raising the level used to dump the whole WebSocket handshake.
Thanks to @Adam-Dalloul, @asteroida123, @codebutler, @dawNotPoi and @lizzjin for contributing to this release.

Google Antigravity joins the built-in agents.
Sub-agent transcripts and task details slide in from the right instead of opening as a dialog you have to close.
New
- Google Antigravity is a built-in agent. Install and launch it from the agent list — history, resume, skills, experts, Office tools, MCP and delegation all wired up — with a settings page for its four sign-in methods. Apple Silicon, Linux and Windows; Google publishes no Intel Mac build.
- Panels slide in beside your work instead of covering it. The mobile sidebar, aux panel, terminal, settings navigation, skill details and task details are inset drawers now, with nothing dimmed behind them.
- Session viewers stack. Sub-agent, work-task and grok child transcripts nest inside one another rather than burying each other, and no longer close when their card scrolls away or you switch tabs.
- Repository items open in a side panel. The full description, every label and the same Start action — without leaving for the forge and losing your filters, page and scroll position.
- Choose which navigation rows the sidebar shows. Automations, To-dos and the Repository panel each switch off in the view options and stay reachable from quick actions.
- A session takes its title as soon as the agent picks one, along with any forum topic bound to it. (#526, @Adam-Dalloul)
- Workspace backgrounds accept GIF, and animated ones play through the blur and the frosted panels.
- Reclaim a finished task's leftover worktree. One button in the task drawer removes the checkout and its work branch; the task stays on the board.
- Tell the agent how to land the merge. The merge dialog gains an optional instructions box — "prefer this branch's side on any conflict", "update the changelog on the way" — and a merge waiting in the queue lands under the instructions it was queued with. (0.28.1)
Improved
- Updated bundled agents. DeepSeek Harness 0.6.0, OpenCode 1.18.21, Cline 3.0.57.
- DeepSeek Harness 0.6.0 support. Image prompts keep their pictures, compaction reports tokens and elapsed time, and forking, image upload and multi-provider models turn on when the agent advertises them.
- The sidebar's view options are sorted out. Toggles move into submenus so Sort by and Section order come first, and expand/collapse-all becomes its own header button that folds the flat sections too.
- System messages show a preview. Claude Code's post-/compact summary was a shut accordion; it renders clamped now, with a toggle only when there is more to see.
Fixed
- Codex sessions created by other tools display properly. One recorded without Codex's usual event channel came back as a bare "Used 25 tools". (#452)
- pi keeps its status lines out of its answers, and routes retries to the shared banner. (#525)
- A proxy typed without http:// no longer breaks agent installs, which used to abort with an unexplained ERR_INVALID_URL; socks5:// addresses are left alone.
- Claude's model picker no longer lists null. Binding a provider with no model wrote literal nulls into ~/.claude/settings.json.
- Office lock files stop previewing themselves, each costing a tab and a watcher over a ~$ file that can never render.
- A GitLab account shows its initials when its avatar can't load, instead of an empty circle.
- The comment written back on an issue stops linking to the wrong page, and a locally merged task no longer reads as work that shipped.
- A repository row keeps its task badge. A stale lookup blanked it, so an issue already being worked on offered Start — and got a second task.
- A finished task's worktree removal isn't refused. Merge recovery left a dead connection behind that read as an agent still working in there.
- A task in review survives a cancelled session. (#546, @lizzjin)
- An agent home written as ~/… resolves to the right place. Hermes built a literal ~ directory beside its launch directory; Antigravity and DeepSeek were refused write access to the tree they actually use.
- A merged task shows "worktree removed" righ

Google Antigravity joins the built-in agents.
Sub-agent transcripts and task details slide in from the right instead of opening as a dialog you have to close.
New
- Google Antigravity is a built-in agent. Install and launch it from the agent list — history, resume, skills, experts, Office tools, MCP and delegation all wired up — with a settings page for its four sign-in methods. Apple Silicon, Linux and Windows; Google publishes no Intel Mac build.
- Panels slide in beside your work instead of covering it. The mobile sidebar, aux panel, terminal, settings navigation, skill details and task details are inset drawers now, with nothing dimmed behind them.
- Session viewers stack. Sub-agent, work-task and grok child transcripts nest inside one another rather than burying each other, and no longer close when their card scrolls away or you switch tabs.
- Repository items open in a side panel. The full description, every label and the same Start action — without leaving for the forge and losing your filters, page and scroll position.
- Choose which navigation rows the sidebar shows. Automations, To-dos and the Repository panel each switch off in the view options and stay reachable from quick actions.
- A session takes its title as soon as the agent picks one, along with any forum topic bound to it. (#526, @Adam-Dalloul)
- Workspace backgrounds accept GIF, and animated ones play through the blur and the frosted panels.
- Reclaim a finished task's leftover worktree. One button in the task drawer removes the checkout and its work branch; the task stays on the board.
Improved
- Updated bundled agents. DeepSeek Harness 0.6.0, OpenCode 1.18.21, Cline 3.0.57.
- DeepSeek Harness 0.6.0 support. Image prompts keep their pictures, compaction reports tokens and elapsed time, and forking, image upload and multi-provider models turn on when the agent advertises them.
- The sidebar's view options are sorted out. Toggles move into submenus so Sort by and Section order come first, and expand/collapse-all becomes its own header button that folds the flat sections too.
- System messages show a preview. Claude Code's post-/compact summary was a shut accordion; it renders clamped now, with a toggle only when there is more to see.
Fixed
- Codex sessions created by other tools display properly. One recorded without Codex's usual event channel came back as a bare "Used 25 tools". (#452)
- pi keeps its status lines out of its answers, and routes retries to the shared banner. (#525)
- A proxy typed without http:// no longer breaks agent installs, which used to abort with an unexplained ERR_INVALID_URL; socks5:// addresses are left alone.
- Claude's model picker no longer lists null. Binding a provider with no model wrote literal nulls into ~/.claude/settings.json.
- Office lock files stop previewing themselves, each costing a tab and a watcher over a ~$ file that can never render.
- A GitLab account shows its initials when its avatar can't load, instead of an empty circle.
- The comment written back on an issue stops linking to the wrong page, and a locally merged task no longer reads as work that shipped.
- A repository row keeps its task badge. A stale lookup blanked it, so an issue already being worked on offered Start — and got a second task.
- A finished task's worktree removal isn't refused. Merge recovery left a dead connection behind that read as an agent still working in there.
- A task in review survives a cancelled session. (#546, @lizzjin)
- An agent home written as ~/… resolves to the right place. Hermes built a literal ~ directory beside its launch directory; Antigravity and DeepSeek were refused write access to the tree they actually use.
Thanks to @Adam-Dalloul and @lizzjin for contributing to this release.

Issues and pull requests can now be handed to an agent without leaving codeg.
Qoder joins the built-in agents.
New
- Issues and pull requests become to-do tasks. A new Repository panel (Beta) lists a folder's GitHub or GitLab items with search, label filters, sorting and paging. Pick one, choose how it should be handled — fix, investigate, plan first, review and fix, or review only — and the agent works in its own worktree while you review the result on the board. Accepting it lands locally or pushes a branch and opens a pull request; a task started from a pull request pushes back to the same branch, forks included. codeg can comment the outcome back on the item, and a task that came from one never auto-merges.
- Qoder is a built-in agent. Alibaba's Qoder CLI installs and launches from the agent list with history, resume, skills, MCP and delegation wired up, and gets its own settings page for the account, access token and Qoder's config file.
- Select text in a message and act on it. A toolbar pops up above the selection to copy it, or quote it into the composer as a Markdown blockquote. Quotes render with a quote rule now, in the composer and in agent messages alike.
- Copy an image out of a conversation. Right-click a picture — in the transcript or the blown-up preview — and pick Copy image. It works in the desktop app now, where copying JPEG, webp and gif used to fail outright.
- Search has a permanent button. Top-left of the window beside the sidebar toggle, and in the mobile title bar, so it stays reachable with the sidebar collapsed instead of only through ⌘K. The remote-workspace picker it replaces moves into the sidebar's right-click menu.
- Start codeg when you sign in. A switch in Settings → System, on macOS, Windows and Linux.
Improved
- Updated bundled agents. OpenCode 1.18.20, Cline 3.0.56, Hermes 0.20.5 and Qoder 1.1.28. Kimi Code is held at 0.36.1 on purpose: 0.37.x stopped bringing up MCP servers handed over on the connection, which takes multi-agent delegation and every user server down with it.
- Quick actions lists every workbench page. The Repository panel joins Automations and To-dos under a Navigation group; Search leaves the menu, now that it has a home that doesn't disappear with the sidebar.
- Updated expert skills. superpowers 6.3.0, plus a tool mapping for Hermes.
- The background-tasks strip is gone. Its count of unresolved background tasks stranded often enough to misinform. Background work still keeps its connection alive.
Fixed
- External links open in the desktop app. Issue titles, delivered pull requests, MCP homepages, OpenCode's docs and links inside previewed Markdown all did nothing when clicked in the packaged app.
- The terminal isn't black in the light theme. On older WebViews the whole terminal painted itself black, because the colour it was handed came in a syntax it couldn't read. (#363)
- Windows paths keep their separators. Markdown was eating the backslash before a punctuation character, so …\.playwright-cli\shot.png rendered — and opened — as a path pointing nowhere. (#508)
- @ works after Chinese, Japanese and Korean text, and on phones. The mention panel only triggered after an ASCII space, which those scripts don't write, and refused to open while a soft keyboard's composition was in progress. It also stays clear of the on-screen keyboard now.
- pi shows what its commands printed. Its bash cards sat on [Terminal: call_…] with no output at all, or painted the raw JSON envelope over the terminal body. (#519)
- A Goal keeps its final answer. Reopening a conversation whose /goal never formally closed folded the whole turn, wrap-up included, into the collapsed capsule.
- Tool results stop showing up as raw JSON. Refreshing mid-turn or opening a session in a second client could leave a tool call painting its envelope instead of its card.
Thanks to @asteroida123 and @Adam-Dalloul for contributing to this release.

Conversations stay where you put them: no panel stuck on "responding", no history disappearing when a new session starts, and no message landing in a session it doesn't own. The status bar also gains a quick-actions menu.
New
- Quick actions live in the status bar. A button in the bottom-left corner opens ten entries that were scattered across the sidebar, folder menus, the window chrome and settings: open a folder, clone a repository, the project launcher and remote workspaces; manage conversations, import local sessions, search; automations and to-dos, carrying the same failure and attention counts the sidebar shows; and the pet. The status bar is always on screen, so all of it stays reachable with the sidebar collapsed. Remote workspaces and the pet are desktop-only.
- Mention a session straight from the sidebar. Right-click a conversation and pick Add to session to drop it into the composer as a mention badge — the same badge the @ panel and the file tree produce. It lands at the cursor, and clicking twice doesn't add it twice.
Improved
- The @ panel lines up with the composer. It now takes the composer's width and edges and opens above it, exactly like the / menu, instead of a narrow box pinned to the cursor. The extra room goes to the description, which used to be cut off while the name sat next to several hundred pixels of nothing. It also stays with the composer it belongs to: a panel left open when you switch to Automations or To-dos no longer floats over the new page, and it keeps its place while the sidebar collapses or a panel is dragged.
- / works while the agent is still connecting. Slash commands only exist once the agent is up, so typing / during connection used to do nothing at all. The panel now opens with a loading row and fills in the moment the commands arrive — no need to retype.
- Updated bundled agents. Hermes 0.20.4, CodeBuddy 2.137.1, Kimi Code 0.37.2, Grok 1.0.5 and DeepSeek Harness 0.5.0.
Fixed
- A conversation no longer gets stuck on "responding" forever. Opening a finished to-do task's conversation from the sidebar left the composer streaming with nothing on the way: Stop did nothing, and only restarting the app cleared it. Two views of one agent — a conversation tab and a task's transcript, or two split tiles — used to take the live connection from one another, leaving the loser waiting on an answer it could never be told had arrived. Both are kept up to date now, and a view that misses the end of a turn checks with the backend and settles itself instead of freezing.
- A new session no longer takes a conversation's history with it. Sending a message could make a long conversation disappear from the sidebar: the agent had quietly opened a fresh session — after a reconnect, or a session that failed to load — and that new session overwrote the one the entire history hung off. The transcript was still on disk, just listed nowhere. The earlier session is now kept as its own entry, with its original name, time and folder, so nothing has to be dug out by hand.
- A message no longer lands in a session another conversation owns. When the agent handed back a session that already belonged to a different conversation, codeg took it anyway and sent the turn, which was then written into that other conversation's transcript — so after a reload your message looked like it had moved. Such a turn is now refused before it is sent, rather than dispatched into the wrong place.
- Codex sessions show the titles Codex gave them. codeg was using the first user message as the title, while Codex keeps the real one in its own session index, so the same session read one way here and another there. The sidebar, the session details and Codex now agree, titles you renamed by hand are still left alone, and the conversation list no longer stalls behind a bound chat channel that is slow to answer.
- Grok reports the tokens it actually spent. The breakdown read "output 0, cache read 0, cache write 0" for every Grok session, because

A follow-up to 0.26.0: retry notices stop piling up under the composer, and a session started from a To-do card keeps the name of the card that launched it.
Fixed
- Retry notices no longer stack under the composer. A long turn that reconnected three times left three amber rows permanently docked over the chat. A warning now clears the moment the agent produces output again, instead of waiting for the turn to end cleanly; several at once collapse into a single strip with a "+N more" count; and every strip — the muted "Recovered" line included — carries a close button. That recovered line also goes on its own after ten seconds, rather than sitting under the composer for the rest of the session announcing a hiccup that was already over.
- A session started from a To-do card keeps the card's name. The name was written at launch and then overwritten on the first detail load by whatever the agent's own session file parsed to — for agents that write no title of their own, the first line of the composed prompt — so the sidebar filled up with rows like 项目：/Users/… that answered to nothing on the board. The name is now locked when the session launches, renaming a card renames the session it produced (a session you renamed by hand keeps your name), and sessions from earlier runs are repaired on upgrade.
- An automation run is named after the automation. Not after the first eighty characters of its prompt, which is identical on every run and never said which automation the session belonged to.

This release is mostly about who is allowed to do what. Two ways a repository could act on your machine without ever asking are closed, an agent can now run its own files and commands so its own sandbox covers them, and OpenCode's permissions get a real editor. Approvals that arrive together no longer hang the agent, and a merge that finds the slot busy queues instead of failing.
New
- OpenCode permissions without the JSON. A Permissions card on OpenCode's settings page: auto-accept everything, set a global default, or pick Allow / Ask / Deny per tool, with pattern rules for the tools that support them. A rule order that would silently override your settings is flagged and fixed in place.
- Let an agent handle its own files and commands. codeg served file reads, writes and terminal commands for every agent, which meant they ran outside whatever sandbox the agent applies to itself. A toggle beside the environment editor (CODEG_ACP_HOST_TOOLS=agent) hands all three back. Off by default — codeg adds no sandbox of its own, so turn it on only for an agent that has one (#436).
- Merges queue instead of failing. Accepting a second reviewed to-do while the project was landing another one used to be refused outright. It now takes a place in line and starts on its own as soon as the slot frees; cards and rows show where it sits and let you leave the queue.
- The connection heart is a button. Click the status icon below the composer for the agent, its real state, working directory, session id and any error — plus a Reconnect that works in every state, including the stuck "connecting" you click it from precisely because it is stuck. It resumes the session rather than starting a new one.
- Live output speed. While an agent is replying, the turn row above the composer shows its output in tokens per second, beside elapsed time and files changed. It also appears in the read-only transcripts on To-dos and loop runs.
- See which agent is on a to-do. Board cards and list rows carry the agent's mark beside the title, resolved the same way the run itself resolves it — so a task inheriting its folder's default shows the agent that will actually run.
- Manage conversations across the whole workspace. The dialog is no longer locked to the folder that opened it: one search row over folder, branch, agent and status filters, with branches folded by prefix so task/49 and task/50 sit under one task/. Each row shows the branch it was started on.
Improved
- Updated bundled agents. OpenCode 1.18.18, Gemini 0.55.1, Cline 3.0.55, Hermes 0.20.1, CodeBuddy 2.136.0, Kimi Code 0.36.1, Grok 1.0.3 and Cursor 2026.08.11.
- The shell you pick applies to agent commands too. The shell setting in General said it covered terminal tabs only; it now also applies when an agent asks codeg to run a whole command line. Leaving it on System default keeps the previous behavior exactly. Agents emit POSIX syntax, so the setting warns that fish, nushell and Windows PowerShell 5.1 may reject it.
- Folder pickers search, and show the real folder name. To-dos, Automations and Token usage now use the same searchable list as the new-conversation composer, showing alias [ name ] over the full path and matching on all three. An alias used to replace the real name, so two folders aliased alike were indistinguishable.
- Worktrees say what they are and where they live. A worktree folder is labeled with the branch checked out there and reads task/49 [ codeg-task-49 ] in the sidebar. Existing worktrees are labeled on startup, and a name you chose yourself is never overwritten.
- The French interface is complete. 61 strings were still surfacing in English — 28 of them the entire Remote Workspace module — and are now translated and aligned to the glossary the rest of the locale uses.
- Release notes come in your language. Releases carry English and Chinese in one body; the settings page and the update popover now render only the half matching your interface language.
Fixed
- pi no longer trusts a

Make codeg yours: recolor any theme, round every corner, or write your own CSS — and let it ring instead of just lighting up a badge. Long conversations open in a fraction of the time, Grok's sub-agents are no longer a black box, and to-do cards accept screenshots. Cline 3.0.50+ works again.
New
- Make any theme your own. A new Custom Style section layers on top of the 12 presets: 31 color tokens (7 up front, 24 under Advanced), light and dark set separately, and one slider for the whole app's corner radius. Themes import and export as shadcn registry:theme JSON through the clipboard, and an opt-in CSS editor covers what the fields can't. If custom CSS ever breaks the interface, ⌘⌥⇧S suspends it and ?safeStyle=1 opens a clean window.
- Notification sounds. Settings → General rings the same five events the chat channels push — turn complete, permission request, agent question, agent error, message sent — each with its own tone, plus volume and an "only when the window isn't focused" gate. Off by default, remembered per device.
- A "Recent" section in the sidebar. Every conversation newest first, 15 at a time, with its own new-conversation button. Folders, Chats and Recent can now be reordered from the view-options menu.
- Grok sub-agents are visible. Each spawn_subagent child gets an Agent card that shows progress while it runs, renders its report as markdown, and opens the child's own session read-only. Child sessions no longer clutter the conversation list.
- Every to-do box is the full composer. New task, edit, follow-up and restart note all get the "+" menu, @ references, slash commands, skills and attachments — so a bug report can be just a screenshot.
- Merge automatically. A folder can land reviewed tasks on its own, exactly as pressing Merge would. Anything that fails stops and waits for you, and is never retried unattended.
- Delete a worktree from the branch selector. "Delete worktree" keeps the branch; "Delete worktree and branch" removes both and moves its sessions to the repo folder. The main working tree, and any tree a task is running in, are never touched.
- An "MCP support" switch for custom agents. Agents that reject codeg's MCP companion couldn't be connected to at all — turn the switch off and they can. A session refused over MCP now points you at it.
Improved
- Long conversations open fast. A conversation loads its last 120 rounds and pages older history as you scroll up, without the view jumping. Server and remote-workspace responses are compressed too; exports still take the whole transcript.
- A to-do whose worktree was deleted is no longer stuck. Reviewed cards offer Complete instead of Merge, every card badges the missing tree, and a retry rebuilds the checkout from its recorded branch instead of stranding committed work.
- The settings pages read as one design. General was rebuilt on the shared card layout, single-option sections shrank to one line, and the five in-conversation agent tools became one "In-conversation tools" card with a single Save button. Custom agent forms follow the same cards.
- Updated bundled agents. Grok 1.0.0, Hermes 0.20.0, OpenCode 1.18.15, Cline 3.0.52, CodeBuddy 2.133.1, Kimi Code 0.34.0. Hermes moves to npm since upstream's PyPI packages stopped at 0.19.0, and its managed install follows HTTP(S)_PROXY. Reconnecting to Grok now resumes the session instead of replaying it.
- Grok's context ring reads the real window. Taken from the wire and Grok's own model catalog rather than guessed from the model name, updated as the turn streams, and re-keyed when you switch models.
Fixed
- Cline 3.0.50 and newer work again. A single unrecognized config option was failing the whole session handshake, leaving the agent unusable (#426). It now appears as a toggle in the composer.
- Enter to pick a CJK candidate no longer fires the button behind it. In WebKit, including the desktop app, that keypress could commit your working tree, create a branch or submit a dialog. About 25 inputs are cov

The Task Board is now To-dos — and it can plan ahead: schedule a card to start on its own, or swap the whole page for a list. Agents can park work too: with your permission, they can save an automation or drop a to-do without leaving the conversation. This release also carries everything from 0.23.4, which never made it out.
New
- Schedule a to-do to start on its own. Pick a date and a time on any To do card; the task stays where it is until then and starts exactly as if you had pressed Start, with the folder's concurrency limit still deciding what actually launches. A time missed while the app was closed fires at the next launch, pressing Start yourself takes over from the plan, and cancelling a task clears it so a later requeue can't resurrect a forgotten time.
- Agents can create automations and to-dos from a conversation. Two new tools let an agent park work instead of only doing it: one saves an Automation (on a schedule or manual), the other queues a to-do card in the right project — worktrees resolve back to the repository they were cut from. Both are off by default under Settings → General → Create from chat, and switching one off stops even already-running sessions from using it.
- A list view for To-dos. One flat, freshest-first list carrying the same actions as the cards — shown as labelled icon buttons instead of hidden behind an overflow menu — plus a status filter that speaks the four column names. The board/list toggle and the settings entry now sit in the window's top-right cluster beside the gear, and your choice of view is remembered.
- Folder badges count what's running. The number beside a folder shows how many of its sessions are working right now — worktrees included — and disappears when none are.
Improved
- The Task Board is called To-dos. The page, the sidebar entry, the settings copy, and the tool description an agent reads all use the new name.
- Custom date ranges got a real calendar. Token Usage's custom range is one pill opening a two-month calendar anchored a month back, so the second pane isn't all-future and greyed out, and clicking into a finished range starts a new one rather than dragging whichever end you didn't touch. Scheduling a to-do uses the same calendar plus a clock, and it opens empty — a task that was never planned no longer looks planned.
- To-dos show a loading skeleton. While the first fetch is in flight the page draws the layout it's about to fill, instead of flashing "no tasks yet".
- The skill shortcuts sit still. The row under the quick actions no longer drifts on its own; an arrow at each end moves it, and each arrow is live exactly while there is somewhere left to go.
- A quieter folder list. The expand/collapse chevron only shows on hover — the folder icon already says whether it's open.
- The rotating tip has caught up. Twelve new tips cover To-dos, Automations, Token Usage, @ mentions, split tabs, worktrees, importing sessions, sub-agent runs, live feedback, Skill Packs, model providers, and backgrounds.
- Tapping Automations or To-dos closes the mobile sidebar, the same way tapping a conversation already did.
- The toolbar's "Process all" button is gone. Start a task from the card itself, or turn on "Process automatically" in Task settings to let a folder work through its to-dos up to the concurrency limit.
Fixed
- Forking no longer renames the conversations it splits. A fork used to lose its [Fork] marker — or quietly revert a conversation you had renamed back to its old name — the next time its details loaded, leaving both halves wearing the same title.
- Date ranges no longer land a day early. The usage range and a scheduled start now speak your local calendar days from end to end, instead of drifting a day for anyone west of Greenwich.
- Deleting a to-do that was just starting cleans up after itself. The delete and the start are now settled against each other, so a run can no longer slip through and leave its worktree and agent process behind.
- Card and row button

In this release, we've focused on making task follow-ups clearer, improving usage reporting accuracy, and polishing several everyday interactions across desktop and web.
New
- Guide reviewed tasks with a clear follow-up intent. Choose Rework, Keep going, Ask, or Double-check when a task needs another pass. Questions instruct the agent to answer without changing files, while a double-check can be started without writing an additional prompt.
- Carry context through retries and cancellations. Retrying or requeueing a task can now include an optional note for the next run. Cancelling or abandoning a task can record a reason in its timeline. These fields use the full composer, including @ references, agent slash commands, and Codex $ skills.
- Complete no-change tasks without merging. When a reviewed task only answered a question or verified existing work, it can be marked complete directly. No empty merge is created, and you can choose whether to keep its worktree.
Improved
- Recent task activity stays at the top. Every task board column now shows the most recently updated cards first, so newly moved, retried, or completed work is easier to find.
- Live guidance reaches Claude more reliably. With the updated Claude Code ACP adapter, instructions sent during a running turn are delivered immediately, and the task remains active until the follow-up work actually finishes.
- ACP troubleshooting logs are more informative. Rejected concurrent prompts and unhandled extension notifications now leave clearer diagnostic traces — thanks to @ijpq for the improvement (#415).
- Bundled agents have been refreshed. Claude Code ACP is updated to 0.65.0, OpenCode to 1.18.14, Cline to 3.0.50, and Kimi Code to 0.33.0.
Fixed
- Token usage now matches what agents report. Claude API calls are counted once, Codex tool-calling rounds are no longer missed, Claude sub-agent usage is included, and usage is assigned to the day the turn began. Existing usage history is rebuilt automatically after upgrading.
- Execution times are accurate across agents. Multi-step Codex turns no longer multiply elapsed time, Gemini no longer includes time spent between messages, and completed Claude replies retain their duration.
- Removed task worktrees disappear immediately. Deleted worktrees no longer remain in the sidebar until a reload, and any local tabs belonging to them are closed.
- Links and popups open reliably in web-based environments. Markdown links and app windows now work correctly in Safari, WebKit-based webviews, and remote desktop mode.
- Hidden editors stay hidden. Git diff panes no longer show through the Task Board, Automations, Token Usage, or conversation overlays.
- Slash command menus are no longer clipped by dialogs. Command suggestions can escape scrollable containers, reposition when space is limited, and remain clickable inside modal views.
- Custom Codex authentication settings are preserved. Saving Codex or model-provider settings no longer overwrites an explicit requires_openai_auth = false configuration.
- Claude transcripts no longer show internal interruption markers. Messages such as [Request interrupted by user], generated internally while applying live guidance, are no longer displayed as if the user had sent them.

New
- A token usage dashboard. The session counter in the status bar now opens a full report: total usage with a previous-period comparison and cache hit rate, day/week/month trends, an activity heatmap, breakdowns by folder, agent and model, and the sessions that consumed the most. Pick any date range, and export the view as a poster card to share.
- Commit, pull and stash from the side panel. The Changes tab commits directly — type a message, press Enter — with pull, fetch, push, stash and the working-tree actions in a menu beside it; quick commits take tracked files only. The Commits tab gets its own menu with pull, fetch, push and manage remotes.
- Update or push any branch without switching to it. Every branch in the picker now offers Update, which fast-forwards it while your working tree stays untouched — and refuses cleanly rather than ever rewriting history — and Push, which opens the push window preselected for that branch so you can review the commits before they go out.
Improved
- The branch menu under the composer slimmed down. Working-tree operations now live in the Changes tab and remote bookkeeping in the Commits tab, so the branch menu is back to picking branches, with only the actions that concern the one you're on.
- Branch lists open instantly in big repositories. The Commits tab's branch filter no longer renders every ref up front, so a repo with hundreds of remote branches opens without the pause — search, prefix groups and keyboard navigation included.
Fixed
- Codex turns that run a list of commands through one script are readable again. They used to reopen as a single card of raw JavaScript with every command's output fused into one blob. Each command now comes back as its own card with its own output and the label the script gave it — and where Codex's log truncation swallowed the boundary between two commands, the card says so instead of showing output that might not be its own.

New
- Multi-folder workspaces. Link other directories into a workspace as subdirectories when you open a folder — a shared library, a sibling service, the docs repo. The agent reads and edits across all of them, and the file tree and workspace search cover them too. Links can be renamed, recreated or removed from the folder menu, and one switch keeps them out of git status.
- Per-stage task prompts. Append your own instructions to any stage a task runs through — first run, retry, return from review, merge. Each stage carries its own example, and your text is added to codeg's built-in prompt rather than replacing it.
- A Preparing status for tasks. Worktree setup and dependency install now show as Preparing under In Progress; Queued means only that a task is waiting for a free slot. Cancelling really stops the install, and an interrupted install runs again instead of starting the agent on half-installed dependencies.
- A way back from full-page views. A back arrow on the Task Board and the Automations page returns you to your conversations, including when the sidebar is collapsed.
Improved
- Task settings redesigned. Split into General, Workflow and Prompts, with options grouped in cards. The merge strategy is two choices side by side instead of a dropdown, and what happens to the working copy afterwards now sits with it.
- One path field everywhere. Home, parent directory and the system picker moved inside the path box, and cloning, new worktrees and new projects all share it. The icon shows which picker opens — the OS dialog for local files, codeg's own browser for files on a server. Choosing a folder fills the field in instead of opening it and skipping the confirm step.
- Task results render as markdown. Headings, lists and code display properly, and a long result folds behind show more.
- Long plans collapse. Plan documents cap their height and offer show more / show less, like long messages already do.
Fixed
- Clicking around the sidebar no longer interrupts a running agent. Replacing a preview tab used to kill the agent mid-turn and leave "Request interrupted by user" in its transcript.
- A new conversation keeps its tab after the first prompt. Tab state from your other devices now merges with what is open here instead of replacing it wholesale.
- Imported sessions keep their /rename title and a truthful timestamp. Claude Code and CodeBuddy both. Re-scanning also refreshes already-imported sessions, so ones you keep using in the agent's own CLI sort by real activity. Renames, pins and status are untouched.

- feat(tasks): New — the Task Board (beta). Write down what needs doing and agents work through the list, each in its own copy of the code, so several run at once without disturbing what you're doing. A finished task waits for you to look it over, then merges when you say yes.
- feat(automations): An automation can now file a task instead of starting a session. A nightly job leaves you a queue to review in the morning, rather than sessions you have to go hunting for. Automations you already saved are unchanged.
- feat(automations): The Automations page has been rebuilt — one toolbar with folder and on/off filters, list and detail sharing a single panel, and Run now and Edit right under the title.
- feat(acp): Every permission option says what it grants, and for how long. Before you press anything, each choice spells out what it would allow and whether it lasts for this call, for the session, or gets written into the repository's settings. Claude Code and Codex both.
- feat(acp): Claude Code and Codex need an adapter, and codeg now says so. They're the only built-in agents whose own CLI doesn't speak codeg's protocol, so having claude on your PATH and still reading "Not installed" was baffling. Their settings card now explains the difference — and that both read the same config directory, so there's no second sign-in.
- feat(codex): Codex asks before acting on a plan, and that request now arrives as a proper plan card instead of an empty one.
- fix(codex): Old Codex conversations read like conversations again. Recent Codex builds pack a whole turn into a single script, which reopened as one card full of raw output; the commands, patches, reads and searches inside get their own cards back. Live conversations were never affected.
- fix(chat): Tool results without a card of their own are readable now — a collapsible tree instead of a blob of JSON, with a switch back to raw text.
- fix(acp): A reply that comes back empty tells you why, instead of one generic line about checking your configuration.
- fix(claude): Typing your own answer to a question keeps working.
- fix(ui): A dropdown inside a dialog no longer closes the dialog with it.
- chore(acp): Updated the built-in agents. Claude Code 0.64.1, Codex 1.1.9, Gemini CLI 0.53.1, OpenCode 1.18.11, Cline 3.0.49, CodeBuddy 2.132.0, Kimi Code 0.31.1, pi 0.0.33 and Grok 0.2.118.

- feat(settings): Kimi Code with your own API key now gets a reasoning picker. A Reasoning card in the panel declares which levels to offer and which one to start on — until now only subscription accounts had the option.
- fix(acp): A long-running command no longer freezes the conversation. Agents that run their commands through the ACP terminal — Grok among them — got stuck halfway through a session with Stop doing nothing, because a command that never exits on its own blocked everything behind it. Such a command now runs in the background with its output streaming as it comes, and the turn stays interruptible. Agents that run commands themselves, Claude Code and Codex included, were never affected (#394).
- fix(codex): Codex searches read like searches. A rg or grep shows as "Grep " with results grouped per file and every line clickable to open it there, instead of a raw JSON blob. Claude, Cline, Gemini, OpenCode and Cursor searches get the same view.
- fix(grok): A command Grok runs in the background finally shows what it did. Each task gets a card with the command, its status, exit code and terminal output, where it used to leave an empty one. Grok's internal notices no longer cut a reply in half either.
- fix(composer): File and command references show as badges wherever text lands in the composer. Quick messages, restored drafts, queued-message edits, expert and Office templates and saved automations used to leave raw [label](file:…) text behind. A Windows path also stops gaining a backslash every time the message is reopened.
- fix(acp): Quitting codeg actually shuts the agents down. Agent CLIs and the processes they start no longer linger after the window closes — thanks to @noxenys for the fix (#389).
- chore(acp): Updated the built-in agents. OpenCode 1.18.10, Cline 3.0.47, CodeBuddy 2.130.0, Kimi Code 0.31.0 and Grok 0.2.114.

- feat(settings): A rebuilt Kimi Code panel with honest status. It reads config.toml back and shows the provider, model and context window really on disk, instead of turning green whenever a token was seeded. Required fields are validated before saving — including "Max context size", which Kimi needs or it discards the whole model block and every prompt comes back empty.
- feat(git-log): Follow the branch you're on in the Commits tab. A HEAD row keeps the history on your current branch across checkouts made anywhere, works while detached, and reports push status against the branch HEAD actually resolves to.
- feat(sidebar): Import local sessions without hunting for the button. The empty sidebar offers it directly, and it's in the right-click menu on the blank area of the list.
- perf(acp): Custom-agent conversations open far faster. Streamed fragments are merged as they're written: a 100-turn conversation goes from 9.3 MB and 78 ms to read, to 0.7 MB and under a millisecond. Existing transcripts read exactly the same.
- fix(chat): Images and files sent from Windows work again. Pasting an image in a web or remote workspace failed with os error 123, and attachments arrived as links the agent couldn't open. Fixed on both sides, so mismatched client and server versions still work (#392).
- fix(acp): Running a script the agent just wrote no longer fails. A "Text file busy" refusal on Linux is now retried briefly instead of surfaced as an error.

- feat(acp): Add any ACP agent to codeg, and use it like a built-in one. Browse the agent registry in Agent settings and install with a click, or add your own by hand with its icon — codeg picks the package that fits your machine and says up front when an agent doesn't support your platform. Once added, it appears in the agent picker and everywhere else an agent is named, its conversations are recorded, listed and searchable, and it takes @ delegation and skills — expert, Office and science packs included. Edit or remove it any time from its settings card.
- feat(agents): An agent you installed yourself is now recognized. If the agent's command is already on your system, codeg finds it and shows its real version instead of leaving it marked "not installed".
- feat(tabs): Split the conversation view into as many panes as you like. Right-click a tab to split right or down, drag the dividers to resize, and drag tabs from one pane to another. Each pane keeps its own tabs and its own new-conversation button, your layout is remembered, and a conversation that's mid-reply keeps running while you rearrange.
- feat(tabs): Unsent drafts come back after a restart. A draft tab returns to the pane you left it in, with the text you had typed still in the box.
- feat(claude): Watch Claude's sub-agents work in real time. A sub-agent's output now streams inside its card as it works, instead of appearing all at once when it finishes. Tool cards and permission prompts also lead with a plain description of what the command does.
- feat(update): New versions announce themselves in the status bar. A badge appears with the version, the release notes and the update button. Dismiss it and it stays quiet in every window until the next release.
- feat(chat): Right-click a file mentioned in a conversation. Reveal it in Finder / Explorer, copy its relative path, or copy its full path — the same actions the file tree offers.
- feat(delegation): Mention a session with @ and the agent will go read it.
- fix(chat): Images sent from the web or a remote workspace no longer disappear. Two screenshots used to be enough for a message to fail silently. Images now upload as you attach them, the limit per image goes from 2 MB to 20 MB, and a failed send tells you and hands your message back.
- fix(composer): Codex skills run properly, and pasted text keeps its references. Picking a Codex skill now sends the form Codex actually understands, and pasting text containing file links or commands turns them back into the badges you'd see in a sent message.
- fix(chat): A link before a Chinese full stop no longer swallows the bracket. (https://xxx.com/o/r)。 now links the address only.
- fix(kimi): Kimi's questions show the answer you picked instead of "no selection" — in existing conversations too, as soon as you reopen them.
- docs(readme): A leaner README, with the full manual at docs.codeg.app. It opens on a workspace screenshot, introduces the iOS and Android apps, and corrects the server install steps.
- chore(acp): Updated the built-in agents. Claude Code 0.63.0, OpenCode 1.18.8, CodeBuddy 2.128.0 and Kimi Code 0.29.2.

- feat(codex): Sandbox and approval policy, from the settings page. The Codex settings card gains a "Sandbox & approval" group: the approval policy (untrusted / on-request / never, or granular with its five switches), the sandbox mode (read-only / workspace-write / full access), and the workspace-write details — writable roots, network access, and the two temp-directory exclusions. These are the thread defaults that /goal, /review, and /compact actually run under, which the composer's collaboration mode never reached: a goal could stay locked down while every ordinary turn in the same session ran unrestricted. Your changes merge into ~/.codex/config.toml with comments and formatting intact, and running Codex sessions are flagged for restart once they change.
- feat(git-log): Long commit messages stay out of the way. An expanded commit in the Commits tab now caps its message body with a bottom fade and a Show more / Show less toggle, so a long release message no longer pushes the file tree and containing branches off the bottom of the panel.
- fix(acp): Agents can reach their own files again. File access over ACP was pinned to the session working directory, so agents that keep state elsewhere failed silently — Grok's plan mode couldn't write its plan and fell back to a shell heredoc. Writes now also cover the agent's own data home and the temp directories, and two new per-agent variables (CODEG_ACP_FS_POLICY, CODEG_ACP_FS_EXTRA_ROOTS) let you add roots or restore the old strict behavior from the existing settings UI.
- fix(delegation): Trustworthy delegation status under Codex. With Codex as the main agent, polling a batch of delegated tasks collapsed into a single anonymous row that printed raw JSON, carried no task id or duration, and showed "done" for tasks that were still running. codeg now unwraps the envelope Codex puts around live MCP results; existing conversations pick this up when you reopen them.
- fix(diff): Closing a diff no longer throws. Opening a file while a diff was on screen — or closing a diff tab, or the commit, push, or stash panels — raised an uncaught editor error. The viewer now tears its editor down in the right order.
- fix(workspace): A cleaner file column. The tab strip keeps its bottom hairline even when a long diff title overflows its tab, maximizing the file column no longer leaves the panel resize handle drawn through it, and the side-by-side diff finally has a crisp divider between its panes instead of just a blurred shadow.
- chore(acp): Refreshed the bundled agents. OpenCode 1.18.5 and Grok 0.2.112.

- feat(claude): Opus 5 support. Claude Code and the model-provider settings now default to claude-opus-5 (and claude-sonnet-5 for Pi), so new sessions target Anthropic's latest models out of the box.
- chore(claude): Updated the Claude Code engine. Bundled Claude Code moves to engine 2.1.219 (claude-agent-acp 0.62.0). Context-window detection now also recognizes the SDK's 1M-lane model id, so those sessions stop falling back to the 200K estimate.
- feat(codex): Answer Codex's prompts with codeg's own cards. Plan-mode questions, MCP forms, and tool-call approvals from Codex now surface as codeg's interactive ask and permission cards — synthesized live and preserved on reload — and its plans render as a proper markdown card. (Codex updated to 1.1.7)
- feat(codex): Pause or clear an active goal. The goal card gains Pause and Clear buttons; resuming is just re-sending your /goal. The controls hide on history reload, for viewers, and in the read-only sub-agent dialog.
- feat(codex): A clearer live sub-agent card. It now shows the sub-agent's model and reasoning effort as it runs, no longer draws an empty white box when there's nothing to report, and picks up Codex's new retry banner and context-compaction card.
- feat(codex): Plan mode moved to the mode selector. The redundant /plan slash command is gone — toggle plan mode from the collaboration-mode selector instead.
- feat(grok): Grok plan mode, end to end. Leaving plan mode now shows an approval card (approve / request changes / abandon) in both the main conversation and the sub-agent dialog, and survives a mid-turn reconnect. "Approve & build" is honored correctly, request-changes notes go back as a follow-up so Grok revises, and plan calls render in the dedicated plan-mode card.
- fix(grok): Long file writes no longer fail. When a command line is too long to run directly, Grok now falls back to the shell instead of erroring with "File name too long."
- feat(grok): Context compaction shown inline. Auto-compaction now appears as a centered divider between turns with the token delta, in both the live stream and history.
- feat(sidebar): A more focused sidebar. Finished conversations are now hidden by default — flip on "Show completed" to see them. Imported local sessions arrive as pending review so they stay visible.
- feat(settings): New Claude Code privacy toggles. Two switches now control Claude Code's attribution/billing header (off by default) and non-essential network traffic (disabled by default), written straight to your native config.
- fix(settings): A cleaner skill import picker. Built-in expert, office, and science skills no longer clutter the agent skill-import list.
- fix(pet): Pet marketplace images load everywhere. Thumbnails and previews are now fetched through the backend, so they show up even where the marketplace host is unreachable from the webview.
- fix(chat): Snappier session switching. The outgoing tab hides immediately, and the new-session welcome controls no longer linger and animate over the conversation you just opened.
- fix(messages): Right-sized message bubbles. Your messages and ask-question results now hug their content instead of stretching to full width.
- feat(layout): Smarter branch search. The branch selector now ranks matches by relevance, with exact matches first.
- chore(experts): Updated expert skills to superpowers 6.2.0. Brings a rules-based "writing good tests" guide, a plan-scoped subagent workspace whose review-fix loop routes back to the original implementer, and forge-agnostic PR creation, among other refinements.
- chore(acp): Refreshed the bundled agents. CodeBuddy 2.127.0, Kimi Code 0.29.1, Pi 0.0.32, Cursor 2026.07.23, and Codex 1.1.7.

- feat(codex): Answer Codex's prompts with codeg's own cards. Plan-mode questions, MCP forms, and tool-call approvals from Codex now surface as codeg's interactive ask and permission cards — synthesized live and preserved on reload — and its plans render as a proper markdown card. (Codex updated to 1.1.7)
- feat(codex): Pause or clear an active goal. The goal card gains Pause and Clear buttons; resuming is just re-sending your /goal. The controls hide on history reload, for viewers, and in the read-only sub-agent dialog.
- feat(codex): A clearer live sub-agent card. It now shows the sub-agent's model and reasoning effort as it runs, no longer draws an empty white box when there's nothing to report, and picks up Codex's new retry banner and context-compaction card.
- feat(codex): Plan mode moved to the mode selector. The redundant /plan slash command is gone — toggle plan mode from the collaboration-mode selector instead.
- feat(grok): Grok plan mode, end to end. Leaving plan mode now shows an approval card (approve / request changes / abandon) in both the main conversation and the sub-agent dialog, and survives a mid-turn reconnect. "Approve & build" is honored correctly, request-changes notes go back as a follow-up so Grok revises, and plan calls render in the dedicated plan-mode card.
- fix(grok): Long file writes no longer fail. When a command line is too long to run directly, Grok now falls back to the shell instead of erroring with "File name too long."
- feat(grok): Context compaction shown inline. Auto-compaction now appears as a centered divider between turns with the token delta, in both the live stream and history.
- feat(sidebar): A more focused sidebar. Finished conversations are now hidden by default — flip on "Show completed" to see them. Imported local sessions arrive as pending review so they stay visible.
- feat(settings): New Claude Code privacy toggles. Two switches now control Claude Code's attribution/billing header (off by default) and non-essential network traffic (disabled by default), written straight to your native config.
- fix(settings): A cleaner skill import picker. Built-in expert, office, and science skills no longer clutter the agent skill-import list.
- fix(pet): Pet marketplace images load everywhere. Thumbnails and previews are now fetched through the backend, so they show up even where the marketplace host is unreachable from the webview.
- fix(chat): Snappier session switching. The outgoing tab hides immediately, and the new-session welcome controls no longer linger and animate over the conversation you just opened.
- fix(messages): Right-sized message bubbles. Your messages and ask-question results now hug their content instead of stretching to full width.
- feat(layout): Smarter branch search. The branch selector now ranks matches by relevance, with exact matches first.
- chore(experts): Updated expert skills to superpowers 6.2.0. Brings a rules-based "writing good tests" guide, a plan-scoped subagent workspace whose review-fix loop routes back to the original implementer, and forge-agnostic PR creation, among other refinements.
- chore(acp): Refreshed the bundled agents. CodeBuddy 2.127.0, Kimi Code 0.29.1, Pi 0.0.32, Cursor 2026.07.23, and Codex 1.1.7.

- feat(git-log): A rebuilt Commit history tab. Virtualized and paged for long histories, with branch and author filters, a pushed/not-pushed marker on every commit, and your filter choices remembered per folder.
- feat(sidebar): Group sessions by git worktree. Turn on "Show worktree folders" to nest each worktree under its repo as a collapsible, branch-labelled sub-group. On by default now, alongside "Show completed."
- feat(delegation): Mention an agent with @ to hand it the work. Naming a sub-agent in your message now delegates that work straight to it — one delegation per agent named.
- feat(chat): View a diff from a reply's changed files. Each modified file now has a View Diff button that opens its diff in an editor tab, on web and desktop.
- feat(layout): Session stats on mobile. The mobile status bar now shows your session stats on the left.
- fix(chat): No more phantom tool cards after a turn ends. Empty tool calls left by interrupted or retried turns no longer resurface and inflate the tool-group and poll cards.
- fix(messages): The focus ring follows your keyboard, not your mouse. Clicking a message no longer leaves a focus outline; keyboard navigation still shows one. Thanks to @youxikexue (#338).
- fix(chat): A tidier connection indicator. It now aligns with the send button and uses a muted colour when connected instead of a green heart.
- fix(layout): No stray scrollbar on the branch selector. The popup now sizes to its rows, so filtered lists have no leftover scrollbar or empty space.
- chore(acp): Refreshed the bundled agent versions. Claude Code 0.61.0, Gemini 0.52.0, CodeBuddy 2.126.0, Kimi Code 0.29.0, Grok 0.2.111, and Cursor 2026.07.20.

- feat(git): The branch selector moves below the composer, with a rebuilt dropdown. It now mounts per conversation, and its dropdown is a searchable, virtualized list combining git operations with prefix-grouped local and remote branches — each offering switch, merge, rebase, and delete.
- feat(chat): Context usage and connection status move below the composer, per conversation. The context gauge and a color-coded connection heart now sit beneath each composer, so tiled views show their own state.
- feat(acp): Diagnose why an installed agent still reads "not installed." A read-only report checks Node/npm, the npm prefix, your PATH, and each agent's launchability, names the likely cause, and copies it all in one click. Open it from Agent Settings or the composer's not-installed banner.
- feat(chat): Slash commands match by fuzzy subsequence. Typing /bmadhelp or /bmhp now finds bmad-help, across both the chat and automation menus. Thanks to @asxuen (#368).
- feat(chat): Changed files in a reply open on click. Changed-file rows now render as cards that open in the workspace tabs, with a reveal-in-folder button.
- feat(sidebar): The locate-conversation button moves to the sidebar header, beside the view-options funnel.
- fix(chat): A copied URL pastes as its address, not the page title. Links copied from a browser's address bar now keep the URL, while text copied within the editor still round-trips intact.
- fix(kimi): Kimi's context gauge no longer jams at 100%. It now reads occupancy from the latest step instead of summing every step's cached prefix; cumulative token totals are unchanged.
- fix(chat): No more stacked placeholder tool cards while streaming. Empty in-flight polls are skipped until a real id or result arrives.
- fix(acp): Install logs keep non-UTF-8 error text, so localized OS errors (e.g. GBK on Chinese Windows) no longer truncate the log or blank the failure detail.
- fix(acp): The composer catches up when an install is detected, refreshing from "not installed" as soon as the Settings probe records a version.
- fix(sidebar): The Folders section stays visible when nothing is open, showing an empty hint instead of disappearing.

- feat(workspace): A reorganized desktop layout. The folder selector now lives in the conversation header — sized to the title and aware of your folder aliases — while the git branch selector and command launcher move into the bottom status bar as paired controls. A one-click pull-code button sits beside the branch name, cloud pull/push actions live in the branch menu, and the redundant pickers below the composer are gone.
- feat(workspace): Mobile now mirrors the desktop. The mobile workspace adopts the same conversation header, file breadcrumb, and bottom status bar under a slimmed top bar. The sidebar and file drawer close on their own after you pick a conversation, start a new one, or open a file.
- feat(import-sessions): Import local sessions in bulk from a dedicated window. A new window scans every local agent store, groups the sessions it finds by project folder, and imports your whole selection in one pass — with live per-agent scan progress, title/path search, an agent filter, and folder- and session-level selection. Open it from a folder's menu or the Folders header.
- feat(chat): Long messages you sent now collapse. A long message you sent clips to a comfortable height with a fading edge and a "Show more" / "Show less" toggle; assistant replies are left untouched.
- fix(chat): Windows file links open instead of reading "[blocked]". Local Windows paths (C:\…, file:///C:/…) — including Chinese, spaced, and URL-encoded ones — now open through the workspace file opener instead of being dropped.
- fix(git): The branch label holds steady after a pull or switch. The bottom status bar no longer flashes "no branch" for a moment after a pull, fetch, commit, or checkout.
- fix(workspace): Cleaner chrome over a background image. The status bar, sidebar, tab-strip line, and active-tab arch keep a legible tinted hairline over a background image, and the new-conversation button's hover and the hovered-tab corners render cleanly.
- fix(logging): Quieter, bounded logs. A debug target that could grow the log file without bound is capped at warnings, and repeated broadcast-lag warnings collapse to one line per window.
- chore(acp): Upgraded the bundled agent versions in the registry. Claude Code 0.60.0, OpenCode 1.18.4, Gemini 0.51.0, Cline 3.0.46, Hermes 0.19.0, CodeBuddy 2.125.0, and Kimi Code 0.28.1.

- feat(chat): Frosted message surfaces over your background image. With a workspace background image on, the surfaces inside chat messages — bubbles, capsules, cards, code headers, tables, and inline code — turn to one family of translucent frosted panels so the image shows through. Inline code also reads as a pill and tables as rounded, bordered cards.
- fix(tabs): The new-conversation button sits neatly beside the tabs. It shares the tab row and stays right next to the last tab, and over a background image its hover frost and the strip's bottom border render cleanly.

- feat(git-log): The Commit history tab is now a continuous timeline. History renders as one virtualized timeline that loads older commits on demand as you scroll, so the whole log stays reachable — not just the first page.
- feat(file-tree): Drive the file tree from the keyboard. Arrow keys move, expand, and collapse the selection, Enter or Space opens a file or toggles a folder, and Home/End jump to the ends — with the selected row kept in view.
- feat(grok): Choose how Grok authenticates. Pick an official subscription (grok login), an XAI_API_KEY, or a custom endpoint; the panel shows only the credential that method needs and recognizes your current one from the saved config.
- feat(workspace): Frosted surfaces over a background image. With a background image enabled, inline code, message bubbles, delegation cards, and Monaco's sticky-scroll header turn to frosted glass instead of opaque blocks.
- feat(workspace): A tidier titlebar at any zoom. The browser-style tabs draw at equal width under a continuous arch, and the window controls no longer overflow when the app is zoomed in.
- feat(workspace): Remote windows show the workspace name. A remote-desktop window's status bar now shows the connected workspace's name — with its address on hover — beside the session count.
- fix(search): The file search dialog finds deeply-nested files. A gitignore-aware backend walk reaches files at any depth without descending into node_modules or target, and fuzzy ranking surfaces deep matches. The composer's @-mention picker benefits too.
- fix(workspace): Cursor and Grok delegation cards show their icon again. They no longer render a blank icon for those sub-agent types.

- feat(acp): Cursor joins as the 12th coding agent. Sign in with your subscription or an API key, pick a model, and use it like any other agent — with full session history, MCP, and delegation.
- fix(chat): Images now work with Grok and other embedded-context agents. Pasted images show as thumbnails and appear in your sent bubbles, and they survive reopening a session.
- perf(chat): Streaming conversations render more smoothly. Several rendering optimizations cut the work repeated on every streamed update.
- fix(files): Editors no longer leak memory. Monaco models are released when file tabs close and when diff editors unmount.
- fix(chat): The new-session composer tracks your agent and focuses on open. It stays in sync with the agent you pick and takes focus the moment a conversation opens, rather than waiting for the ACP connection.
- fix(commands): A command's run state survives reopening the sidebar. Closing and reopening the sidebar keeps the command's run state and your selection instead of resetting them.
- fix(workspace): Background-image refinements. The image now shows through the right sidebar header, and the opacity sliders reach further — the mask up to 99% and panels down to a fully transparent 0%.

- feat(chat-channel): Telegram Topic mode. Run multiple sessions inside one Telegram forum supergroup, each bound to its own topic. /task <description> opens a new topic and session, /resume <id> binds an existing one, and plain text follows up in the current topic. Requires a forum supergroup with the bot as a topic-managing admin. Thanks @AnotiaWang (#322).
- feat(codex): Custom Codex models without hand-editing JSON. A structured editor in Codex settings lets you define custom models and curate the list — clone an official model, override only what differs, pick enum fields from dropdowns. Official models are read live from the launched Codex, so new ones appear automatically.
- feat(grok): Real Grok permission modes and reasoning effort. Permission mode now uses Grok's own --permission-mode values (old settings migrate automatically), and the reasoning-effort selector follows the chosen model — showing only the efforts it supports, or hiding entirely when it has none.
- feat(skills): Author your own skills. A new Custom tab in Skill Packs lets you create, edit, duplicate, and delete skills in the shared store and enable them for any agent — write once, reuse everywhere. Import from a local folder or from an agent's existing global skills.
- fix(chat): Async sub-agent results settle cleanly into their turn. Held-turn content no longer double-renders, the launch card updates in place to its finished result, and the "syncing results" strip stops lingering.
- fix(chat): Correct first-reply stats after resuming a conversation. The first reply after reopening a conversation no longer absorbs every earlier turn's duration and tokens — each reply reports only its own.
- fix(acp): A session's last error survives a reconnect. Reconnecting after missing the live error event still shows what went wrong, and a stale error is never resurrected onto a session that moved on. Thanks @ijry (#314).
- fix(chat-channel): Telegram tokens no longer leak; @username channels bind correctly. Bot tokens are scrubbed from error messages, and an @username is resolved to its numeric chat id so a follow-up no longer spawns a duplicate agent.
- fix(session): Completed sessions show their true runtime, computed from start and end timestamps. Thanks @cnYui (#311).
- fix(editor): Windows IME composition is preserved when typing Chinese, Japanese, or Korean in the built-in editor. Thanks @youxikexue (#332).
- fix(docker): Docker builds include the pnpm workspace config, so container and self-hosted deployments build cleanly. Thanks @AnotiaWang (#322).
- chore(agents): Bundled agents refreshed — Claude Code (0.59.0), OpenCode (1.17.20), OpenClaw (2026.7.1), Cline (3.0.40), CodeBuddy (2.121.2), Kimi Code (0.24.1), Grok (0.2.101).

- feat(settings): Experts, Science, and Office Tools are now one "Skill Packs" hub. The three separate settings pages are merged into a single Skill Packs entry with a tabbed switcher for Experts, Scientific Research, and Office — plus a fixed toolbar to open the shared central skills folder or refresh the active tab. Old links to the individual pages redirect here automatically.
- feat(chat): Kick off Scientific Research skills straight from the composer. The composer's "+" menu gains a "Scientific Research" submenu next to Experts and Office Work, listing every bundled science skill; pick one to drop its invocation badge into your message. Skills you haven't enabled for the current agent show a lock and a shortcut to turn them on in Settings.
- feat(layout): Roomier sidebars. The left and right sidebars can now be dragged out to 900px, giving conversation lists and side panels more breathing room on wide displays.
- fix(chat): Quick-skills now use the right prefix for whichever agent you send to. A skill badge inserted before switching agents kept its original trigger, so a skill queued for Codex went out as /skill and Codex rejected it as a slash command. Badges are now re-stamped when you change agents and normalized at send time — $ for Codex, / everywhere else — so the invocation always lands.
- fix(acp): A conversation that can't be resumed now offers a clear way forward. When a historical session can't be restored — the agent process crashed mid-load, or the session has expired or ended — codeg shows a Reload / New conversation banner instead of silently orphaning the context or surfacing a raw protocol error.
- fix(editor): No more phantom errors or orange boxes in the file editor. The built-in editor no longer flags false-positive TypeScript/JSON problems that only exist because it can't see your project's build context ("Cannot find module '@/…'", a missing --jsx flag, unreachable schemas), and Chinese/Japanese full-width punctuation (：；，！？（）) renders as normal text instead of highlighted boxes. Real syntax errors and zero-width/BOM detection are still flagged.
- fix(editor): Markdown list markers show up in file preview. Ordered lists now display their numbers and unordered lists get proper indentation in the editor's Markdown preview.
- fix(chat): The model picker no longer closes when you grab its scrollbar. On the desktop app's WebKit webview, dragging the model list's scrollbar used to dismiss the picker; it now stays open in both the wide picker and the collapsed cog panel.

- feat(grok): Bring your own Grok endpoint. Grok Build settings gain a "Custom model (BYO endpoint)" section — point it at any Grok-compatible endpoint with your own model id, base URL, API key, backend, context window, and session auto-compact threshold. Leave the URL blank to fall back to xAI's official endpoint.
- feat(grok): Grok no longer updates itself in the background. The Grok CLI launches with its self-updater disabled, so it keeps running the pinned build instead of silently swapping to a newer one — and its interactive approval prompt stays intact unless you've switched to always-approve.
- feat(grok): The context-window meter now works for Grok sessions. Grok conversations show how much of the model's context window you've used, with the window sized per Grok model (from 256K up to 2M) and paired with Grok's running token count — just like every other agent.
- fix(icon): Monochrome agent icons no longer look dimmed. The OpenCode, Hermes, Grok, Cline, and CodeBuddy marks now render at full strength in the agent selector, sidebar, and status bar, instead of appearing faint until you hover over or select them.
- chore(agents): Bundled agents refreshed. CodeBuddy Code (2.118.2) and Kimi Code (0.23.5) were bumped to their latest builds.

- feat(acp): Meet Grok Build — the 11th built-in agent. xAI's Grok Build joins codeg as a first-class coding agent: sign in with your xAI API key, choose its model and reasoning effort, delegate sub-agents, and manage its MCP servers — with live and replayed sessions rendered exactly like every other agent.
- feat(science): A new Scientific Research skills mode. Thirteen curated, open-source research skills (from K-Dense-AI's scientific-agent-skills) now ship alongside the bundled expert and office skills. Install them to the shared library, enable them per agent in Settings, and start from the new "Scientific Research" tab on the new-conversation screen — localized in all 10 languages.
- feat(chat): The composer and your sent messages are now plain text. What you type, and the messages you've already sent, display as clean plain text with file and session references shown as compact badges — so your input reads exactly as written instead of being reflowed as Markdown. (Assistant replies are unchanged and still render Markdown.)
- fix(chat): Background tasks now settle correctly. A background task that's been collected or stopped is now marked as finished, so its indicator no longer lingers as if it were still running.
- fix(chat): The model picker stays open while you use its scrollbar. Dragging the scrollbar inside the model list no longer dismisses the picker.

- fix(mcp): MCP servers you add now save correctly for every agent. Because each agent expects its MCP config in a different format, an entry codeg wrote could previously be silently ignored — or make an agent reject its whole MCP config and disable all of its servers at once. codeg now writes each agent's config in the exact format it accepts, so your servers stay intact across every agent.
- feat(models): You can now run Codex with OpenAI's new GPT-5.6 series. The refreshed Codex build adds support for the GPT-5.6 models, and they're offered as examples in the model field so they're easy to pick.
- chore(agents): Bundled agents refreshed. Claude Code and Codex were updated to their latest builds, and OpenCode (1.17.18), Gemini (0.50.0), Cline (3.0.39), and Kimi Code (0.23.3) were bumped alongside them.

- feat(workspace): Showing and hiding the sidebar, side panel, and terminal now animates. These panels slide open and closed instead of snapping in place, so rearranging your layout feels smoother.
- fix(worktree): Worktree conversations are listed in full, and their Git status stays live. The conversation manager now includes sessions nested in subfolders, and a worktree's Git status refreshes on its own as you commit, switch branches, or move HEAD — no manual reload needed.
- fix(preview): In-page anchor links now work in the HTML preview. Clicking a link that jumps to a #section on the same page scrolls to that heading inside the sandboxed preview, instead of doing nothing.
- fix(chat): The folderless "chat mode" entry no longer scrolls out of reach. In the folder picker it's now pinned as a sticky footer, so it stays visible and selectable even under a long list of folders or an active search filter.
- fix(ui): A few title-bar and indicator alignment touch-ups. The macOS traffic-light buttons are now vertically centered in the title bar, and the background-tasks indicator strip sits properly centered.
- chore(agents): Bundled agents updated — OpenCode 1.17.15, Hermes 0.18.2, CodeBuddy 2.117.2, and Kimi Code 0.23.2.

- feat(agents): Claude's built-in sub-agent activity now shows up live. Task runs, background shell watchers, and hooks that fire mid-reply appear as they happen, instead of only surfacing after you reopen the conversation.
- fix(fork): Forking a conversation is reliable again. Forks no longer fail with a "database is locked" error, and forking a resumed conversation now branches from the right point instead of replaying duplicated history.
- fix(sidebar): Delegation sub-sessions are ordered newest-first. The latest sub-agent appears at the top under its parent, matching the rest of the sidebar — no more scrolling to the bottom to find it.
- fix(i18n): The live tool-use counter now appears in your language. The per-turn tool-activity count shown while a reply is streaming was previously left untranslated — it's now localized across all supported languages.
- perf(workspace): The first code block in a reply renders without a stall. The syntax-highlighting engine now warms up off the streaming path, so it's ready before the first code block appears.
- perf(workspace): Cosmetic conversation updates skip the stats recompute. Changes that only touch a conversation's status, title, or pin state no longer trigger a token/message-count recalculation.
- chore(agents): Bundled agents updated — Claude Code 0.57.0(0.3.202), Gemini 0.49.0, OpenCode 1.17.14, CodeBuddy 2.117.1, Kimi Code 0.23.1, and Cline 3.0.38.

- perf(streaming): Streaming replies do far less work per token. Only the turn being written is recomputed, so long, fast responses stay smooth.
- perf(streaming): The conversation view no longer re-renders on every streamed token. It updates only from the fields it displays.
- perf(sessions): A session streaming in the background no longer slows the one you're watching. State moved to fine-grained stores that update only the affected panels.
- perf(agents): The installed-agent list is fetched once and shared instead of rescanned per component on every window focus.
- perf(git): The Git panel opens instantly on huge change sets. Very large change lists cap at 500 rows with a one-click reveal.
- perf(diff): Enormous diffs open without locking the UI. Previews render the first 500 lines with a "show the rest" button.
- perf(history): Conversation and folder lists load faster as history grows. Per-file summaries are cached instead of re-parsed on every listing.
- perf(startup): Text-only sessions load lighter and faster. The highlighting, math, and diagram engines (~10 MB) load only when actually used.
- perf(acp): Lower overhead while events stream. Event sizes are estimated structurally instead of serialized under the session lock.
- feat(editor): Long lines can now wrap in the editors. Toggle word wrap (off by default) from the right-click menu, Alt+Z, or Fonts settings.
- feat(composer): Paste as plain text with Ctrl/Cmd+Shift+V, skipping the usual Markdown/rich-text conversion.
- feat(appearance): Hide the mode cards on the new-conversation screen with a new Appearance setting (on by default).
- fix(chat): The per-reply files card stays compact — created and changed files now share one bounded, scrollable area.
- chore(experts): Bundled expert skills refreshed to Superpowers 6.1.1.
- chore(acp): Bundled agents updated — CodeBuddy 2.117.0 and Kimi Code 0.22.3.

- feat(workspace): Open file tabs now stay put when you switch folders. Switching the active folder no longer closes your open files, and files from different folders can stay open at the same time.
- feat(chat): Every completed reply now shows the files it produced. An inline artifacts card above the turn stats lists that reply's new and changed files — click a new file to open it in the editor (or reveal it in your OS file manager), and expand a changed file to view its diff inline.
- feat(codex): Codex /goal now renders as a dedicated goal card, in both live sessions and reopened history, instead of plain text.
- fix(files): File links are more reliable, including Windows network shares (UNC paths). UNC and protocol-relative links now open in the right place — handed to the file opener instead of the browser — with correct casing and canonical URIs.
- fix(composer): Quick messages containing tables or task lists now insert correctly. Picking a saved quick message with a table or a - [ ] checklist no longer silently does nothing.
- fix(desktop): Fixed a lingering double title bar on Linux after upgrading. Window decorations are now decided per platform at launch and no longer restored from saved window state, so the native title bar won't reappear over the app's own toolbar.
- chore(acp): Bundled agents updated — Claude Code 0.55.0, Codex 1.1.0 (codex 0.142.5), OpenCode 1.17.13, Hermes 0.18.0, CodeBuddy 2.114.2, and Kimi Code 0.22.1.

- feat(conversations): Delegated sub-sessions now show as a live tree in the sidebar. Expand a chat to see and open its sub-sessions.
- feat(sidebar): You can now choose whether Folders or Chat appears first in the sidebar.
- fix(chat): Expanding an edited file now shows just the changed part instead of the whole file.
- fix(desktop): Fixed a double title bar on Linux windows.
- fix(feedback): Fixed Live Feedback breaking some slash commands.
- chore(acp): Updated the bundled agents and added Claude Sonnet 5 support.

- fix(chat): Plan mode now renders as a dedicated card. Claude Code's plan steps and Cline's mode switches show their plan markdown in a proper plan card instead of folding into a generic "thinking" group.
- fix(chat): Claude Code background tasks now collapse into one lifecycle card. A background command's launch, its repeated status polls, and its stop merge into a single card with the final status, exit code, and ANSI-decoded output — instead of a new raw card per poll.
- fix(acp): Kimi Code permission requests now show readable text. Prompts that carried only a description (and no structured fields) used to render as raw JSON; they now display the request text.
- fix(codex): Codex thinking no longer fragments on reload. Consecutive reasoning steps regroup into a single thinking block when a session is reopened, matching the live stream.
- fix(chat): Mermaid diagrams stay crisp when zoomed. Zoomed diagrams no longer rasterize into a blurry image on macOS.
- fix(agents): Clearer error when a Windows agent upgrade hits a file lock. Upgrading or uninstalling an agent while its process is still running now tells you to close its sessions and retry, instead of showing a raw "could not remove directory" error.
- release(macOS): macOS builds are signed and notarized again. Re-enabled Apple notarization for desktop releases.

- feat(feedback): Live Feedback now keeps the agent checking your notes on its own. Once enabled, the agent periodically reads your mid-task notes and adjusts course — no need to remind it each turn.
- fix(chat): Codex file edits now render as proper edit cards, with their diff and line-change stats instead of a generic tool entry.
- fix(chat): Kimi Code task lists now render as a plan card, matching Claude Code, in both live streaming and history.
- fix(acp): Codex permission prompts now show clearer request details — the relevant tool title, command/input, and working directory.
- chore(acp): Bundled agents updated — Claude Code 0.53.0, Codex 1.0.2 (codex 0.142.4), Cline 3.0.33, CodeBuddy 2.114.0, and Kimi Code 0.20.2.

- feat(pi): Added first-class Pi agent support with pi-acp integration, session history parsing, delegation, provider setup, credentials, thinking level, and a dedicated settings panel.
- feat(pi): Pi runtime can now be managed from Codeg with install/uninstall actions, live logs, workspace trust setup, and clearer missing-runtime prompts before launch.
- feat(pi): Pi now works with Codeg-managed skills, experts, and Office Tools when using the default managed directory.
- feat(composer): Added Experts and Office shortcuts to the "+" menu, with locked-state hints and safe Office template insertion.
- polish(pi): Improved Pi provider names, provider coverage, and agent icon styling.

- feat(chat): Codex sub-agent collaboration now renders as clean, collapsible capsules — each sub-agent shows its task, live status, and result as markdown instead of raw spawn/wait/close tool calls, both live and when you reopen the session from history.
- feat(composer): A flowing gradient border marks the active session's composer in tiled multi-session workspaces, replacing the old top-center icon. It shows only while the input isn't focused; focusing it restores the standard focus ring.
- fix(office): Office Tools (officecli) now work in Docker and server deployments — the runtime images were missing the ICU library officecli needs, which had silently broken skill sync and document preview there. Such failures are now surfaced clearly instead of passing as success.
- chore(acp): Bundled agents updated — Codex on codex-acp 1.0.1 (codex 0.142.2), Cline 3.0.31, CodeBuddy 2.112.1, and Kimi Code 0.20.1.

- feat(acp): Connect AI providers to OpenCode directly from its agent settings. Browse the models.dev catalog and bring a well-known provider online with just an API key, or define your own custom OpenAI-compatible provider. Connected providers are organized into catalog and custom lists — each with enable/disable, edit, and disconnect — and the model picker now surfaces context-window and reasoning metadata. Keys are written only to OpenCode's own auth.json.
- feat(office): Office Tools (officecli) now install on Windows, not just macOS and Linux. Installation runs the vendor's official script (mirror-first, with a GitHub fallback) and streams its output live into the Office Tools settings page, so you can watch progress instead of waiting on a spinner.
- fix(office): Made the officecli installer far more robust — the whole install is now bounded by a timeout (and kills the entire process tree, including the background binary download, if it expires), a stalled mirror fails over to GitHub instead of hanging, and downloads on older, locked-down, or non-English Windows hosts now succeed and log correctly.
- fix(office): Live document preview for agent-generated .docx/.xlsx/.pptx files now opens automatically whenever a workspace folder is open — even with the file side panel closed.
- fix(db): Fixed an intermittent "no such column" migration failure that could surface under load, by applying schema migrations on a dedicated connection before the runtime read pool opens.

- feat(office): Office documents are now a first-class workflow — create, analyze, proofread, and edit .docx, .xlsx, and .pptx files through the bundled officecli toolset. A new welcome-page quick-actions panel (Coding and Office tabs) drops the matching skill invocation and a prompt template into the composer with one click.
- feat(office): Preview .docx/.xlsx/.pptx right in a file tab — the preview opens automatically and refreshes live as the agent edits, backed by a long-lived officecli watch server (reverse-proxied and capability-authenticated for web and server deployments).
- feat(settings): A new Office Tools settings page installs officecli and manages its document skills.
- feat(settings): The Experts and Office Tools pages now use a skill-by-agent matrix — toggle any (skill, agent) pair, enable or disable a skill across all agents or every skill for one agent, and apply bulk changes to selected agents at once.
- feat(chat): Expert skills now appear directly in the "/" and "$" slash menus alongside the agent's other commands, instead of in a separate submenu.
- feat(chat): Welcome-page quick actions now respect per-agent skill enablement — a skill that isn't enabled for the selected agent shows a lock badge and links you to where you can turn it on.
- feat(chat): The welcome composer is taller and its conversation column is wider for more comfortable first prompts.
- feat(acp): Agents that support it now restore historical sessions via session/resume (falling back to session/load, then session/new), so reopened conversations resume faithfully.
- fix(acp): npx-distributed agents now install their platform-specific native binaries even when npm is configured to omit optional dependencies, fixing a "native binary not found" launch failure (notably on Windows).
- fix(acp): Resumed Codex sessions now stay on the model provider configured in ~/.codex/config.toml, instead of being forced back to OpenAI's official endpoint.
- fix(chat): The composer stays fully focusable while the agent is still connecting.
- fix(chat): Truncated @-mention suggestions now reveal their full text on hover.
- chore(acp): Bundled agents updated — Claude Code 0.52.0, OpenCode 1.17.11, Cline 3.0.30, CodeBuddy 2.111.0, and Kimi Code 0.20.0.

- feat(agents): Kimi Code (Moonshot) is now a built-in agent — sign in with an API key or OAuth, browse and validate available models, manage its MCP servers and skills, and reopen past Kimi sessions from history.
- feat(agents): CodeBuddy (Tencent) is now a built-in agent — with its own sign-in panel, full session history, skills, and MCP delegation. Private, self-hosted CodeBuddy deployments are supported through a custom endpoint URL.
- fix(render): CodeBuddy and Kimi Code sub-agents and delegations now display correctly — both as they run and when you reopen a past session — and CodeBuddy's internal sub-agent transcripts no longer clutter the history list.
- fix(render): Tool-result images — such as reading an image or a multi-page PDF — now appear inline when you reopen a session from history, matching the live view.
- fix(render): Codex live ask-a-question prompts and file-read commands (cat/sed/head) now render as proper cards instead of raw tool output.
- fix(acp): Agents and the terminals they open now run in the session's working directory (with a shell fallback), so file paths resolve as expected.
- fix(branches): The branch picker below the input now navigates worktree branches for existing conversations, instead of refusing or hijacking a checkout.
- fix(automations): Starting a session from the Automations page now returns you to the workspace, instead of opening it hidden behind the full-page Automations view.
- fix(acp): Enabling a codeg-mcp feature (delegation, feedback, ask, or sessions) no longer breaks OpenClaw session startup — codeg now skips the MCP servers OpenClaw can't accept.
- fix(chat): Composer selector tooltips now name the setting they control (config, model, mode) on hover.
- chore(acp): Bundled agents updated — Claude Code 0.50.0, OpenClaw 2026.6.10, and Hermes 0.17.0, with Codex now running on codex-acp 1.0.0; new agents ship at CodeBuddy 2.109.3 and Kimi Code 0.19.1.

- feat(automations): Save a composer setup as a reusable Automation that runs headlessly — on a cron schedule or on demand.
- feat(logging): A new Settings → Runtime Logs page shows the app's logs live, with filtering and per-module levels.
- feat(commands): The command manager is now a searchable, two-pane dialog with drag-to-reorder, inline editing, and per-command delete.
- feat(branches): Branches that share a prefix now fold into expandable groups across all three branch selectors, keeping long lists tidy.
- feat(experts): Open the central skills folder directly in your file manager.
- fix(settings): Reveal the app's ~/.codeg directories straight from the file manager.
- fix(settings): The Settings window is now independent — it no longer moves or minimizes with the main window (#295).
- fix(git): A worktree on a detached HEAD is now recognized as a git repository (#279).
- fix(delegation): A sub-agent's ask_user_question prompt now appears in the session dialog, so you can answer it.
- fix(chat): The ask-question card stays in the message flow instead of overlaying earlier messages (#285).
- chore(acp): Bundled agents updated — Claude Code 0.49.0, Gemini CLI 0.47.0, OpenCode 1.17.9, OpenClaw 2026.6.9, and Cline 3.0.29.

- feat(composer): The model selector now groups models by provider and opens long lists in a searchable, virtualized picker.
- fix(composer): On narrow layouts, picking an option from the collapsed selectors now actually applies it instead of doing nothing.
- fix(ui): Long text in confirmation dialogs now wraps instead of overflowing.

- fix(chat): The session's model selector now shows each model's full name in a tooltip on hover — thanks to @H1d3rOne (#288).
- fix(install): The install script now works for root users and on systems without sudo (containers, slim images), and a half-finished install self-heals on re-run instead of staying broken.
- fix(experts): Bundled expert-skill scripts now keep their executable bit when extracted, so skills that invoke them by path no longer fail.
- chore(acp): Bundled Claude Code updated to 0.47.0.
- chore(experts): Bundled expert skills (superpowers) updated to v6.0.2.

- feat(appearance): The interface font now defaults to Inter.
- feat(claude): A new custom model option — set your own model id, name and description to surface a custom entry (for example a model behind a gateway or proxy) in Claude Code's in-session model picker, available both per agent and on Claude model providers.
- fix(composer): Pasted file paths and filenames (lib.rs, notes.md, setup.io, …) are no longer turned into clickable links.
- fix(message): Copying a user message now copies its literal text instead of escaped Markdown — Windows paths and other backslashes are kept verbatim.

- feat(composer): A full right-click menu — Cut, Copy, Paste, Select All and Quick Messages — anywhere in the message input.
- feat(composer): Attached files and selections now insert at the caret instead of the end.
- feat(composer): The @ mention panel lists only enabled agents.
- feat(editor): "Add Selection to Chat" adds the current line when nothing is selected, and a new "Add File to Chat" adds the whole file.
- feat(conversations): A new Session Details dialog shows a session's metadata and token usage.
- chore(acp): Bundled OpenCode updated to 1.17.7 and Gemini CLI to 0.46.0.

- feat(composer): The message input is now a rich-text composer — type @ for a mention panel that drops inline badges for files, sessions, commits, agents, and skills, with Markdown rendered as you type.
- feat(chat): Attached files now show as inline, clickable badges in your sent messages.
- feat(editor): Select lines in a file and add them to the conversation as a line-range badge.
- feat(mcp): A new get_session_info tool resolves an @-mentioned session into its metadata, stats, and recent messages.
- feat(appearance): The interface font now defaults to System Monospace.
- fix(conversations): File links in conversation titles now fold to their label instead of raw Markdown.

- feat(sidebar): Open Folder and Clone Repository are now one hover away on the Folders header.
- fix(tabs): Quitting with only an unsent draft open no longer reopens to a blank, unusable screen.
- fix(chat): Empty scratch folders left by unsent or deleted chats are now cleaned up at startup.

- feat(chat): Start a quick conversation without picking a project folder — a new "Chat" section sits at the top of the sidebar. While you're in a chat, the folder/branch pickers and side panel step aside.
- feat(sidebar): Start a new conversation right from any folder header, with a consistent icon everywhere.
- feat(chat): The message navigator is now a single chip next to the plan and sub-agent chips — open it to jump between your messages and see each one's changed files and diff.
- feat(tabs): Middle-click a tab to close it — in the session, file, and terminal tab bars.
- fix(settings): The Settings menu now lines up neatly and scrolls when it's taller than the window.
- style(chat): The composer's "+" button is toned down to match the selectors next to it.

- feat(sidebar): A reworked sidebar with a fixed New chat / Search bar (with shortcut hints) and a dedicated collapsible "Pinned" section up top. Hover any conversation for quick pin/unpin and mark-done actions, and your pins and collapsed state are now remembered between sessions.
- feat(chat): The message thread is now keyboard-scrollable — click into a transcript and use Arrow keys, PageUp/PageDown, Home/End and Space, just like the conversation list.
- feat(channels): When an agent asks you a question, that prompt is now pushed to your IM channels and webhooks. On by default, never debounced, and localized across all 10 languages.
- feat(hermes): A smoother Hermes setup — uv runtime install is now its own streamed step with an "Install uv" preflight action, and every uvx call pins Python 3.13 to avoid install failures on newer Python.
- fix(update): In-place server upgrades no longer fail on Docker's overlayfs; the web-asset swap now falls back to a copy when a directory rename is rejected.

- feat(delegation): Hermes can now take on delegated work in multi-agent collaboration. Following its debut in 0.15.5, Hermes is now a selectable target for codeg's sub-agent delegation — so when one agent hands off a self-contained sub-task, that work can be routed to Hermes right alongside Claude Code, Codex, OpenCode, Gemini, OpenClaw, and Cline.
- chore(registry): Bundled Claude Code updated to 0.44.0, which brings support for Anthropic's new Claude Fable 5 model.
- style(chat): The folder and branch picker row now lines up with the composer controls, keeping the folder icon aligned with the "+" button across UI zoom levels.

- feat(agent): Hermes Agent joins the workbench. Hermes (from Nous Research) is now a first-class agent alongside Claude Code, Codex, OpenCode, Gemini, OpenClaw and Cline. A guided setup in Settings walks you through choosing a model provider and entering your key — with a linked editor for the key and the Hermes config — and codeg provisions the Python toolchain for you on first launch. Hermes works with more than 30 providers out of the box (OpenRouter, OpenAI-compatible endpoints, Anthropic, Google AI Studio, DeepSeek, xAI Grok, Z.AI / GLM, MiniMax, Kimi / Moonshot, Qwen, GitHub Copilot, Ollama Cloud, Nous Portal, and more), including your own custom OpenAI-compatible endpoint. It also supports MCP servers — so codeg's multi-agent delegation and your own MCP tools are available — and loads your existing Hermes conversation history into the sidebar.
- feat(chat): Questions you've answered now stay in the conversation. Building on the ask-a-question card from 0.15.3, an answered question is kept as a dedicated card in the message stream, mirroring the live card's layout with your chosen options shown as selected. The record stays collapsed by default to keep the transcript tidy, and expands whenever you want to revisit what was asked and what you picked.
- feat(chat): Agent plans now render as an inline card. When an agent shares or updates a plan, it appears as a dedicated card right in the conversation instead of only behind the corner chip. Long plans cap their height and scroll, so a large checklist never takes over the view.
- feat(chat): Live feedback checks read as a tidy capsule. When an agent picks up the notes you've left through Live feedback, the result now shows as a collapsed capsule that summarizes how many notes arrived and previews the first one, expanding to the full list with locale-aware send times. Routine checks that found nothing new — and checks still in flight — are hidden so they don't clutter the transcript.
- chore(registry): Bundled Claude Code updated to 0.43.0.
- style(chat): The composer toolbar controls are more compact.

- fix(appearance): The interface now defaults to the System UI font rather than a monospace face, and the editor font is confined to the code editor. Choosing an editor font no longer reaches the rest of the app — conversation text, file paths, tool-call parameters, command menus, and shortcut chips follow the interface font and a stable system monospace instead.
- fix(chat): A finished turn's execution-time tooltip now matches the live timer above the composer, using the same localized "Xh Ym Zs" format instead of a separate "1.5m" / "30.0s" style.

- feat(AskUserQuestion): Agents can now pause and ask you a question before moving on. When an agent needs a decision, it raises a card above the composer with up to four single- or multiple-choice questions — each with a "Recommended" hint and an always-available "Other" field — and waits for your answer; multiple questions are laid out as tabs, you can answer from any open window, or skip to let the agent use its own judgment. On by default, and can be turned off in settings.
- feat(conversations): Conversation titles now read like a summary instead of the opening prompt. Codeg adopts the title each agent generates for itself (Claude, Gemini, OpenCode, …) and keeps it in sync as the session grows. Renaming a conversation locks its title so auto-titling never overwrites your choice, and re-importing a folder refreshes the titles it brings in.
- feat(appearance): Customize fonts for the interface, the code editor, and the terminal independently — each with its own size, a ligature toggle, and a choice of bundled or custom font families.
- feat(settings): Settings changes now reach conversations that are already running. After you save — for example, editing a model provider — each affected running conversation shows a banner to reconnect and pick up the new config; reverting the change clears it, and unrelated saves stay quiet.
- feat(chat): The plan and sub-agent overlays in the top-right corner now rest as a compact circular chip instead of a wide pill, so they no longer crowd the conversation; hover or focus it for the full summary, click to expand.
- style(sidebar): Running and cancelled conversations now show a clear status icon in place of a text badge.
- fix(terminal): Fast typing in the terminal stays in order and no longer drops characters under rapid input.
- chore(registry): Bundled Codex updated to codex-acp 0.16.0.

- feat(steering): Nudge an agent while it's working — send a quick note from the composer's "+" menu and it's picked up at the next checkpoint. Off by default.
- feat(update): Updates now download in the background and don't stop when you switch pages — restart when you're ready.
- feat(pet): The desktop pet shows your running sessions at a glance; open the panel to approve permissions or jump into a conversation.
- feat(worktree): Git worktrees are now first-class — create one to open a session in its folder, grouped under its root repo, and switch branches without checkout conflicts.
- feat(chat): A navigator on the conversation's right edge marks every message — click to jump, expand to see and open each turn's file changes.
- feat(chat): Your own messages now render as Markdown.
- feat(web-service): See every local address, with copy and QR-code buttons to open the app on your phone.
- feat(directory-browser): The directory picker now follows the path input — type or click, then Select.

- feat(backup): Export a portable backup of your codeg data and restore from it later. Backups can be passphrase-encrypted, optionally bundle your agent CLI transcripts (Claude, Codex, Gemini, …), and restoring previews the archive and checks version compatibility before doing anything. Your current data is snapshotted first, so a restore can be rolled back if needed.
- feat(file-preview): Added a sandboxed HTML file preview that renders documents safely with scripts disabled, and the preview header now shows the HTML document's title.
- feat(sync): Open conversation tabs and the focused tab now stay in sync across your desktop and browser clients.
- feat(chat): The most recent reply's delegated sub-agents now surface in a top-right overlay, so you can follow delegated work at a glance.
- feat(web): Losing the WebSocket connection now recovers in place through a reconnect dialog, instead of forcing you to log in again.
- feat(chat): Markdown links and file paths in chat now show a type icon in front of them for quicker scanning.
- perf(sidebar): The conversation list is now virtualized for smooth scrolling and rendering even with a large number of conversations; a status change re-renders only the affected card, and sticky folder headers return as a floating overlay.
- fix(chat): Pasting an image from the clipboard now works on Linux and inside the Tauri webview.
- fix(sync): Viewing a shared conversation mid-turn no longer shows a duplicated in-flight prompt or reply.
- chore(registry): Bundled agents updated — Claude Code 0.42.0, OpenCode 1.16.2, Gemini CLI 0.45.2, and OpenClaw 2026.6.1.

- feat(sync): Desktop-hosted web clients now stay in real time with the desktop app. When you open a running conversation from another client, Codeg joins the existing agent stream instead of starting a separate one, and sidebars update as conversations are created, renamed, deleted, or enter review/running states.
- feat(chat-channel): Message Channel events can now be delivered to Webhook URLs as structured JSON, with per-URL enable switches, a management dialog, and safer failure logging.
- feat(chat-channel): Channels can notify you when an agent is waiting for permission. User-message notifications are also available as an opt-in event with bounded prompt previews, and remain disabled by default.
- feat(file-tree): The file tree context menu can now copy paths for the workspace root, folders, and files.
- fix(sync): Shared conversations now queue overlapping sends instead of dropping them, preserve send order across desktop and browser clients, and avoid duplicate user prompts when joining an in-progress turn.
- fix(chat-channel): Event delivery now fails closed when event-filter settings cannot be trusted, Lark cards render untrusted fields as plain text, and failed settings loads no longer overwrite existing filters or webhooks.
- fix(clipboard): Copy actions now work in non-secure web contexts, including components that call navigator.clipboard.writeText directly.
- chore(format): Added Prettier ignore coverage and normalized formatting.

- feat(server): Server and Docker deployments can now upgrade themselves in-place — no manual file replacement needed. The supervisor automatically validates the new version and rolls back if the trial run fails.
- feat(server): A manual rollback action is now available in Server Update settings, letting you revert to the previous version at any time.
- feat(delegation): Sub-agent sessions now open in a centered dialog, making it easier to follow a delegated task's progress alongside your main conversation.
- feat(delegation): Batch-polling delegation status is now supported — query multiple task IDs in a single call for lower overhead in multi-agent workflows.
- feat(tabs): Press Ctrl+Tab / Ctrl+Shift+Tab to cycle forward and backward through open conversation tabs — thanks to @mvanhorn (#246).
- fix(server): The access token no longer regenerates on restart, so existing clients stay connected after a server reboot.
- fix(delegation): Token usage is now correctly backfilled on the child sub-session reply, so cost tracking stays accurate.
- fix(model-provider): API keys containing non-ASCII characters no longer cause a panic when the key is masked for display.
- fix(update): The server upgrade process is now crash-safe: the web bundle swaps atomically, disk writes are verified, and a bad upgrade is self-healed on the next boot.
- fix(plan): The plan overlay now expands correctly, and stale plans are no longer shown where they should be hidden.
- chore(registry): Bundled Claude Code updated to 0.41.0 and Gemini CLI to 0.45.0.

- feat(delegation): Checking a delegation's status with the wait time set to 0 now waits until the task finishes, so you get the result without polling.
- perf(message): Streaming or switching one session no longer re-renders the others — multi-session layouts stay smooth.
- fix(message): The streaming sub-agent card's "Sub-agent starting…" placeholder is now localized across all 10 languages.
- fix(message): Ordered markdown lists now render their markers, and markers stay on the same line as the item text.
- fix(message-input): Clipboard images can now be pasted into the input on Linux.
- fix(goal): Codex goal updates now render as active cards.
- fix(delegation): Reloaded conversations now show the correct delegation status and grouping.

- feat(delegation): Streamlined long-running delegation updates — repeated status checks are folded into one compact card per task, making busy multi-agent conversations easier to scan.
- fix(delegation): Completed “still running” status checks now settle into a neutral checked state instead of looking like live spinners forever.
- fix(parser): Claude Code slash commands that expand into prompts, such as /init, now appear as user turns so the surrounding assistant replies stay correctly separated.
- docs(community): Refreshed the WeChat QR code images for both light and dark documentation assets.

- feat(delegation): Sub-agent delegations now run in the background — the main agent keeps working while multiple tasks run in parallel, and they survive reload and restart.
- feat(delegation): Delegation tool calls now show as compact inline cards with live status and a one-click link to open the sub-agent's conversation.
- feat(delegation): Added a configurable memory budget for delegation results in multi-agent settings (default 512 MB).
- feat(hyperframes-launcher): New HyperFrames tab in Project Boot scaffolds an HTML-to-video project and opens it in the workspace.
- fix(delegation): Delegation status now stays reliable across desktop, web, and server modes.
- fix(settings): Switching the Codex provider no longer applies the wrong settings.

- fix(delegation): Hardened sub-agent delegation under parallelism and cancellation — parallel calls bind to their own tool calls, and cancels no longer mis-bind or strand delegations.
- fix(shadcn-launcher): Scaffolded projects now open in the workspace and the launcher closes itself.
- fix(shadcn-launcher): Preview loading overlay no longer sticks forever.
- chore(workspace): Lowered workspace window minimum size to 400x600.

- feat(acp): Install agents at a custom version — pin any version for NPX and binary agents, or switch an installed agent to a different one.
- chore(acp): Bumped the bundled Claude Code CLI to 0.39.0.

- feat(models): Added support for Claude Opus 4.8 — the latest Anthropic flagship model is now available across Claude Code sessions and delegation flows.
- fix(workspace): Resize handles between adjacent panels are now disabled when the neighboring panel is collapsed, preventing accidental drags on hidden splitters.
- chore(acp): Bumped bundled agent CLI versions — Claude Code 0.38.0, OpenCode 1.15.12, Gemini CLI 0.44.1.

- fix(delegation): "View sub-agent conversation" button no longer goes missing under tool-call id races.
- fix(message): Removed React duplicate-key warnings inside grouped tool calls.
- chore(delegation): Default max delegation depth lowered from 2 to 1.

- feat(delegation): View sub-agent conversations in a read-only side panel — works on both live and reloaded conversations.
- feat(delegation): Multi-agent collaboration is now opt-in — enable it in Settings to use delegate_to_agent.
- perf(workspace): Faster external change detection — overhead now scales with the number of changed files instead of the number of open tabs.
- fix(workspace): Switching between file tabs no longer flashes a loading state or risks clobbering unsaved edits.
- fix(workspace): External edits to background file tabs are detected and reloaded automatically.
- fix(workspace): Externally deleted files now show an explicit error instead of silently displaying stale content.
- fix(workspace): Eliminated races where the watcher could overwrite unsaved edits or paint stale diff markers.
- fix(workspace): Resolved git lock contention between the file watcher and AI agents — thanks to @Kwensiu (#215).
- fix(workspace): Opening a conversation from the sidebar now releases the maximized file pane overlay.
- fix(export): Desktop conversation export now uses the system Save As dialog and surfaces real save errors.
- fix(delegation): Sub-agent errors (auth required, refusal, token limits) now appear in the delegate_to_agent result instead of a generic "canceled".
- fix(delegation): Fixed parent-side hang when a sub-agent connection errors out.
- refactor(conversation): Moved the "Reload" entry below "Export" in the conversation context menu.

- refactor(workspace): Workspace mode now switches automatically between conversation and fusion based on open file tabs, removing the manual mode selector and stale persisted files mode.
- feat(chat): Message links now open directly from chat, with hardened web/desktop handling for external, mailto:,

, and file-path links.
- fix(pet): Permission responses now clear the waiting animation immediately across local windows, WebSocket reconnects, server-mode peers, and chat-channel auto-approval.
- feat(acp): Added Claude Code ACP compatibility handling with normalized tool calls and updated UI classifications.
- chore(acp): Bumped bundled agent registry versions — claude-agent-acp 0.37.0, Gemini CLI 0.43.0, OpenClaw 2026.5.20, Cline 3.0.9, and OpenCode 1.15.7.
- chore(rust): Formatted backend sources.

- Improved the new conversation flow so default agents are resolved more reliably across folder defaults, available-agent ordering, restored draft tabs, and retry/new-session entry points.
- Refreshed the welcome panel with a cleaner centered layout, softer background treatment, highlighted greeting text, and a lighter rotating tip area.
- Added a full test pipeline covering frontend lint/Vitest/build plus Rust desktop/server checks across Linux, macOS, and Windows.
- Expanded Rust and frontend coverage with parser snapshots, HTTP/WebSocket integration tests, command core tests, and React component regression tests.
- Consolidated web handlers around shared command core functions so desktop and server entry points exercise the same business logic.
- Updated README, AGENTS, CLAUDE, and translated docs with the new test and verification commands.
- Cleaned up cross-platform Rust warnings and CI edge cases, including Windows desktop test handling and stricter clippy coverage.

- feat(folder): Right-click a folder → "Set default agent" to pin which agent opens for new conversations from that folder.
- fix(tab): Auto-replaced draft tabs now inherit the previous agent instead of the global default.
- chore(deps): Dropped @lobehub/icons for inlined brand SVGs — smaller install, faster cold start.

- feat(chat): Skill $ autocomplete now shows a one-line description, and the default skill template ships with name/description frontmatter.
- feat(chat): Folder and branch pickers moved to a borderless row below the message input.
- style(chat): Polished the welcome screen — softer elliptical halo, theme-tuned rings, tighter spacing, no more horizontal scrollbar in tiled mode.
- style(settings): Swapped the order of Model Providers and Quick Messages in the settings nav.
- fix(settings): Skill id is locked while editing an existing skill to prevent accidental rename.
- fix(acp): OpenCode sub-agent (task) calls render as Agent cards during live streaming, matching historical reload.
- fix(acp): Saving a skill now returns the latest description so the UI no longer shows a stale value.
- fix(settings): OpenCode enabled_providers allowlist is honored and synced, and provider.<id>.npm is backfilled with the first option.
- fix(model-provider): Codex overrides OPENAI_MODEL in env_json so the selected model takes effect at launch.
- chore(acp): Bumped claude-agent-acp to 0.36.1 and OpenCode to 1.15.5.

- feat(model-provider): Each provider is now bound to a single agent type and carries its own model selection — five slots for Claude Code, one for Codex and Gemini. Switching providers cascades the URL, key, and model into the agent in one step.
- feat(chat): Refreshed the chat welcome screen with a new halo hero and rotating tips.
- fix(model-provider): Agent type is now locked after creation, and mismatched bindings fail loudly instead of silently mis-parsing the model.

- feat(workspace): Web and remote workspaces now support uploading and downloading files and folders directly from the file tree, with streamed transfers, no size caps, a unified progress dialog with cancel support, and hardened safety checks for uploads and zip downloads.
- fix(acp): Custom base URLs configured for the Codex provider are now properly applied.
- fix(settings): The Claude Opus model placeholder now shows the correct value.

- feat(chat): Web and remote workspaces now support file attachments.
- fix(uploads): Uploads now enforce safer size limits, optional storage quotas, path checks, and clearer errors.
- fix(remote): Remote desktop attachment uploads and WebSocket traffic now stay routed through the selected proxy connection.
- fix(acp): ACP sessions now recover more reliably after refreshes and WebSocket reconnects.
- fix(acp): Restored conversations now keep saved selector preferences.
- fix(server): Data directories are now resolved consistently for databases, uploads, and pet assets.
- feat(debug): Event metrics are now available for replay, snapshot, and subscription lag diagnostics.
- chore(acp): Clippy now passes without the connection manager warning.

- fix(remote): all remote-workspace HTTP and event traffic now goes through the desktop proxy, so window navigation, directory pickers, and API calls stay scoped to the selected connection.
- fix(remote): project-boot windows opened from a remote workspace inherit the remote context, so create/clone flows route through the proxy.
- fix(remote): WebSocket subscriptions are now isolated per window instance, so closing one window can't disturb a sibling that reused the same webview label.
- fix(settings): system network settings no longer try to load local state inside a remote workspace.
- fix(pet): pending-review and permission-request cues play as one-shots and return to idle, instead of pinning a "reviewing" ambient state.
- style(sidebar): the active conversation title uses regular weight.

- fix(transport): Web mode no longer hangs on the "Connecting…" state when an ACP agent initializes before the WebSocket subscription is ready.
- fix(transport): mid-session WebSocket reconnects now restore ACP state automatically, so open conversations keep running without a manual reload.
- fix(acp): switching the agent or session at the same workspace tab no longer lets a stale snapshot overwrite the new connection's state.
- fix(transport): closing or switching a remote workspace fully releases its WebSocket, avoiding zombie reconnects and unexpected login redirects.
- fix(install): the install scripts now remove shadowing codeg-server binaries by default (--no-cleanup to opt out) and exit non-zero when the new install isn't what runs.
- fix(install): the Windows installer also cleans shadows on first install when its destination is freshly appended to PATH.

- feat(remote workspace): remote Codeg workspaces can now be saved, tested, reordered, and opened from the desktop app with token-authenticated HTTP and WebSocket transport.
- feat(sidebar): the sidebar header now includes a quick new-conversation action for the active workspace.
- feat(sidebar): themed conversation groups now use app theme colors, clearer folder states, and improved open-tab highlighting.
- chore(experts): bundled Superpowers expert skills have been synced to v5.1.0.
- chore(acp): Gemini CLI has been updated to 0.42.0 and OpenCode ACP has been updated to 1.14.50.

- fix(web): command authentication failures now return as command errors without invalidating the Web session, so valid browser sessions stay signed in.
- fix(db): Tauri debug builds now use an isolated codeg-dev.db, avoiding SQLite conflicts with installed release builds during local development.
- chore(pet): pet sprite animation timing has been aligned across states for smoother idle, action, and review animations.

- feat(web): the built-in Web service can now start automatically when Codeg launches, using the saved port and token. Thanks to @pia for #161.
- fix(tabs): workspace and file tabs now use long-press drag on touch devices, reducing accidental switches, closes, and context menu triggers while reordering. Thanks to @pia for #162.
- feat(pet): the pet manager and marketplace now show animated action previews, so a pet's major states can be checked before activation or installation.
- fix(pet): marketplace search now uses the correct query parameter, so keyword filtering returns the expected online results.
- fix(window): dialog-style windows such as Settings, Commit, Merge, and Push now stay scoped to their opener instead of floating above the entire desktop.
- feat(window): dialog window titles now follow the active app language across all supported locales.
- feat(ui): dialogs now include a consistent close button, including the command dialog.
- fix(acp): overlapping ACP connection attempts now preserve the latest request, so rapid agent or session switches no longer get overwritten by an older connection.
- fix(acp): Gemini ACP now skips the trust prompt for a smoother connection flow.
- chore(acp): OpenCode ACP has been updated to 1.14.48.
- fix(gemini): Gemini session history now supports newer JSONL logs, including incremental message updates, tool result summaries, and token usage.
- fix(chat): token statistics now use compact K/M formatting consistently across message details and the status bar.

- fix(pet): the desktop pet now reliably stays in the Running state across conversation switches, so the animation no longer gets pinned on Review after a previous turn finished or granted a permission.
- fix(pet): opening the pet window mid-conversation now immediately reflects the agent's current state instead of stalling on an outdated idle pose.
- feat(pet): changing the active pet in Appearance Settings instantly hot-swaps the sprite in the live pet window — no need to close and reopen for the new look to apply.
- tune(pet): the jumping animation has been retimed to follow a natural gravity arc, giving the hangtime a more believable, weighty feel.

- feat(pet): meet your new desktop companion — a Codex-compatible pet that lives on your desktop, reacts to chat events with rich animations, and is fully managed from Appearance Settings.
- feat(pet): browse, install, and switch pet characters from the new built-in marketplace with one click.
- feat(pet): hover to get a wave, drag to flip the pet's facing direction, and enjoy looped idle and event animations — all without ever stealing keyboard focus from what you're working on.
- feat(pet): right-click the pet for a native menu to control it without opening Settings.
- feat(chat): see exactly how long the assistant took to reply, right in the message footer.
- feat(codex): images generated by Codex now render inline as first-class content blocks instead of plain links.
- feat(window): closing the main window keeps codeg running in the tray so background sessions and notifications continue uninterrupted; quit fully from the tray menu.
- feat(window): on macOS, the tray icon now uses a monochrome template that adapts to light and dark menu bars.
- feat(window): the Settings window now stays attached to its owner, so it can no longer get lost behind other windows.
- fix(updater): the in-app updater reports the actual installed version when running in server mode, so update prompts are accurate.
- fix(acp): agents that run git commands now pick up your local git credential helper, so private repositories work without re-authenticating.
- chore(acp): bump Codex ACP to 0.14.0.
- style(conversations): tiled conversation panels can shrink to 24rem, fitting more sessions side-by-side on smaller screens.
- i18n(chat): grouped tool runs (read/edit) now show call counts instead of file counts, reflecting actual activity more accurately.

- feat(sidebar): add an "open in" submenu and full-path tooltip to folder groups.
- style(sidebar): replace the conversation status SVG with a unified colored-dot badge shared by the sidebar, tab bar, manage and search dialogs.
- style(sidebar): show the agent icon at the rail axis with the conversation status as a corner badge in each sidebar row.
- style(chat): tint streaming shimmers (tool-group capsule, Agent tool-call pill, running and reasoning indicators) with the theme primary color.
- chore(acp): bump claude-agent-acp to 0.33.1 and opencode to 1.14.41.

- feat(chat): group consecutive tool runs into a compact summary pill with expandable details.
- feat(chat): align multi-turn usage, duration, and model stats with historical reloads.
- fix(chat): preserve file:// links in assistant markdown so local file links still reach the open-file dialog.
- fix(mcp): accept streamableHttp and other casing/separator aliases consistently across manual config, Codex TOML, marketplace, and Smithery paths, normalizing them to the canonical http type.
- fix(mcp): localize unsupported MCP type and transport errors, and route catch-handler stringification through toErrorMessage so backend error payloads no longer render as [object Object].
- test(mcp): cover MCP type normalization across Codex TOML and marketplace paths.
- chore(acp): bump OpenCode, Gemini, OpenClaw, and Cline registry versions.
- style(chat): refine tool call card accents with theme-colored borders and a simpler visual treatment.
- style(conversation): show a focus indicator on the active conversation tile.

- feat(login): localize the web login page across all 10 supported languages.
- fix(acp): resolve npx-launched ACP agents from the npm prefix, and cache the resolution so agent listing no longer stalls on repeated lookups. Thanks to @Kwensiu for #151.
- fix(chat): resume the original ACP session after a reconnect on new-conversation tabs, instead of starting a fresh session.
- fix(chat): mark conversations as cancelled (not failed) when the user cancels a streaming turn, including the sidebar badge label.
- fix(chat): surface agent turn failures with an explicit error state instead of leaving the turn silently pending review.
- fix(chat): show the streaming thinking indicator while reasoning is in flight and hide the expand toggle when reasoning is empty.
- fix(chat): correct the Claude Code agent icon hover title.

- feat(chat): render Codex-generated images inline with a download action.
- feat(settings): require a proxy URL when enabling the system proxy.
- fix(chat): stop the assistant reply from duplicating when a turn completes.
- fix(chat): stop drag-dropped files from leaking into inactive conversation tabs.
- fix(chat): surface session load failures with reload and new-conversation actions.

- fix(mcp): forward configured MCP servers to ACP sessions, with PATH resolution for stdio and capability filtering for HTTP/SSE.
- fix(mcp): write OpenCode MCP entries to the correct config file (~/.config/opencode/opencode.json).
- feat(settings): add refresh and new-MCP draft form to the local MCP tab.
- refactor(acp): move conversation external_id and fork row writes to the backend.
- fix(chat): preserve final stream chunk, jump to bottom on first open, wrap live turn stats on narrow panes, and clamp session config dropdown to viewport.

- refactor(chat): reorganize the message input toolbar — folder/branch pickers move into the top context bar, expert skills and slash commands collapse under the (+) menu, and the bottom toolbar gains an inline session config + mode selector (collapsed under 34rem).
- chore(acp): bump bundled agent versions — claude-agent-acp 0.32.0, codex-acp 0.13.0, @google/gemini-cli 0.40.1, opencode 1.14.33.
- fix(chat): keep the message input scrolled to the caret when content overflows.

- fix(sidebar): keep cancelled conversations visible by default; only completed is filtered out unless "show completed" is toggled.
- feat(chat): auto-collapse the agent tool-call card body once when it transitions from running to completed (non-error). Errors stay expanded; manual re-expand still works.

- feat(chat): consolidate the per-session selectors (model, mode, agent settings) into a single cog dropdown with hover submenus on the message input action row, with the agent icon shown on the trigger and the dropdown labelled as "agent settings"; submenu widths are capped so long option lists no longer push the row off-screen. Localized across all 10 languages.
- feat(chat): move the folder and branch pickers out of the conversation context bar and into the message input action row, so workspace controls stay grouped with the input area where they're actually used.
- style(chat): widen the minimum tile width for tiled conversation tabs to 28rem so multi-tab layouts have more breathing room before content wraps.
- fix(acp): move conversation cancel-on-disconnect from the frontend to the backend lifecycle subscriber via a CAS update (in_progress → cancelled), so a concurrent user-driven completed or a prior pending_review from TurnComplete can no longer be silently overwritten. Eliminates the related frontend races (flash-disconnect during reconnect, dbConvIdRef dependency gap, post-unmount in-flight writes).
- fix(acp): raise the default CODEG_ACP_IDLE_TIMEOUT_SECS from 60s to 180s, giving the frontend keepalive (30s cadence) a 6× safety margin against network jitter and brief stalls before the backend reaps a connection.

- feat(terminal): configurable default shell with backend-driven picker — get_available_terminal_shells / probe_terminal_shell_path serve the platform-aware option list, install state and resolve_shell() fallback; the settings UI renders the response verbatim and supports a custom shell path on every platform. POSIX spawn now distinguishes bash-family shells from unknown shells like nu/xonsh/elvish so the latter spawn cleanly without bash-only -l -i / eval flags. Localized across all 10 languages. Thanks to @Kwensiu for contributing the initial implementation in #127.
- feat(chat): file paths in Read/Write tool call headers and Edit/ApplyPatch diff headers are now clickable and open the file in the workspace panel via the existing local-file confirmation dialog; new-file Write/NotebookEdit bodies render with the same green tint as added diff lines.
- feat(chat): cap inline code block height inside messages with internal overflow scroll, so long code dumps no longer push the rest of the conversation off-screen.
- fix(dialog): constrain DialogContent to viewport height with internal scrolling, and cap the git reset commit message at 8rem with its own scrollbar, so the dialog footer stays visible on long messages or short viewports.

- feat(acp): persist in-flight session state across tab close/reopen, browser refresh, and reconnects. The streaming reply, active tool calls, pending permission prompts, plan updates, modes and usage all rehydrate from a backend snapshot, with seq-based deduplication against the live event stream.
- feat(acp): harden the connection lifecycle with an idle sweep, spawn deduplication mutex, serialized prompt sends, and a close-error state machine; sends are cancelled on failure and stuck spawns no longer wedge the session.
- feat(acp): keep open conversation tabs alive via a frontend keepalive so connections aren't reaped while the user is still looking at them.
- feat(acp): create the conversation row on the first prompt and emit a conversation_linked event, eliminating duplicate rows when the same prompt is sent twice in quick succession.
- fix(web-server): release the listening port reliably on shutdown and surface stale-port state instead of silently failing to restart.
- fix(web): fall back to a clipboard shim on non-secure contexts so copy actions work over plain HTTP.
- fix(workspace-state): apply the cold-scan snapshot after stream restarts and catch up missed deltas when re-acquiring within the shutdown grace window.
- fix(mobile-web): improve mobile web UI control accessibility. Thanks to @pia for contributing this fix in #122.

- feat(settings): add a Windows toggle to disable WebView2 hardware acceleration, helping users work around white-screen and rendering glitches on certain GPUs.
- fix(acp): uninstall before reinstall when upgrading npx-managed agents to avoid stale binaries.
- fix(acp): bypass Windows file locks when clearing the binary cache so re-installation no longer fails.
- fix(acp): dedupe slash command results at the session update mapping layer to prevent duplicated entries in the slash menu. Thanks to @MoozLee for contributing this fix in #111.

- fix(chat-input): raise toolbar collapse threshold to 34rem so selectors stay visible at wider input widths.
- fix(acp): prevent Windows terminal hangs caused by orphaned pipes and .cmd shims.
- fix(acp): fix context loss when resuming a historical conversation by deferring auto-connect until the session id resolves.
- fix(sidebar): clarify empty state when completed conversations are hidden.
- chore(acp): bump claude-code to 0.31.0, codex to 0.12.0, opencode to 1.14.23, gemini to 0.39.1.

- feat(settings): add quick messages management with drag-and-drop sorting and place nav below agents.
- feat(message-input): turn plus button into menu with attach files and quick messages entries.
- feat(codex-skills): surface built-in .system skills as read-only.
- feat(sidebar-conversation-list): show empty-state hint inside expanded folders with no visible conversations.
- fix(sidebar): restrict folder reorder drag to the folder header row.
- fix(message-input): persist draft across restarts by keying on conversation id.
- fix(acp): prevent memory blowup from streaming terminal tool output.
- fix(web-server): release port reliably on stop by awaiting graceful shutdown.
- fix(experts): recognize Windows junction links to central store.
- perf(workspace-state): compress delta history, cache gitignore lookups, and tighten debounce.

- feat(sidebar): add per-folder color swatch with color picker and neutral conversation rail.
- feat(conversation-context-bar): unify selectors and attachments into a horizontal scrolling row.
- fix(conversation-context-bar): hide selectors divider when no attachments are present.
- fix(conversation-manage-dialog): sort conversations by created_at and display creation time.
- fix(icons): regenerate multi-resolution Windows ICO for crisp desktop icon.
- chore(acp-registry): bump opencode to 1.14.21 and gemini-cli to 0.39.0.

- feat(workspace): migrate from per-folder windows to a unified single-window workspace.
- feat(sidebar): overhaul sidebar with draggable folder reorder, per-folder new/import actions, conversation sort toggle, completed filter, and bulk-action management dialog.
- feat(acp-agent-settings): expand OpenCode provider.npm options with additional AI SDK providers.
- fix(transport): replace Tauri listen() to avoid unlisten race crashing on WKWebView.

- feat(message-input): add search box to slash-command popups.
- fix(file-tree): keep aux-panel tree in sync with filesystem changes.
- fix(file-tree): invalidate both cached target and ancestor, skip collapsed refetches.
- chore(acp): bump claude-agent-acp to 0.30.0 and opencode to 1.14.19.

- feat(terminal): add word/line cursor shortcuts for shell line-editing.
- feat(appearance): apply UI zoom level to terminal and Monaco editors.
- fix(settings): move content padding inside scroll areas to keep inner borders visible.
- refactor(git-error): drop locale-specific not-a-repo patterns.
- fix(process): apply C.UTF-8 locale across platforms for child processes.

- feat(web-service): allow custom access token with persisted port and localized start errors.
- feat(conversation): add copy text context menu action with preserved selection.
- feat(status-bar): compact large token counts with K/M units in usage popover.
- feat(chat-channel): annotate Weixin channel badge with iLink reconnect hint tooltip.
- fix(chat-channel): detect Weixin auth failure via errcode during connect verification.
- chore(acp): bump claude-agent-acp to 0.29.2, opencode to 1.4.11, gemini-cli to 0.38.2, openclaw to 2026.4.15 and cline to 2.15.0.