Kelivo: vanhemmat versiot ja versiohistoria

What's Changed
- feat: Add automatic context compaction, with editable summary templates, per-conversation settings, and undo. OpenAI native compaction can be enabled separately.
- feat: Add a fixed context start, so history is sent from a chosen message without rolling truncation by message count.
- feat: Add Standard, Fast, and UltraFast response speed controls in the reasoning picker, saved per conversation and model.
- feat: Add a per-conversation control for image recall, plus local budgets for image size, image count, and request body.
- feat: Detect vision, tools, and thinking support on local Ollama models.
- feat: Let each assistant choose which memory types are always injected in full.
- feat: Add custom model icons, including color and monochrome LobeHub styles.
- feat: Add an optional legacy code block style.
- feat: Quote selected reasoning text into the composer.
- feat: Render Markdown and LaTeX in in-app translation results.
- feat: Add a custom balance query, with its own path, method, headers, and body.
- feat: Add Russian, Korean, and Turkish.
- fix: Limit a browser agent pause to the conversation that started it.
- fix: Recover orphaned Android workspace processes.
- fix: Preserve Codex sessions and Claude thinking across retries and tool continuations, and avoid signing out when a usage check fails.
- fix: Keep follow-up tool requests aligned with the current round after a provider or model change.
- fix: Limit Mermaid rendering memory, and close Windows WebViews before the plugin is torn down.
- fix: Preserve the iOS browser scroll position when the safe area changes.
- fix: Show full model names in lists and settings.
- fix: Keep desktop send shortcuts available in narrow windows.
- fix: Show reasoning activity while the summary is still empty, and restore context usage estimates for chats with images.
- fix: Allow keyboard suggestions on ordinary provider fields while keeping API keys on the secure keyboard.
- fix: Fix several other known bugs and UI issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.3.4...v1.3.5

What's Changed
- feat: Add a desktop side panel that keeps files, topics, and browser tabs together for each conversation, with shortcuts.
- feat: Adapt desktop layouts down to a 360px-wide window.
- feat: Add isolated workspace web apps. A .app.html file keeps its own saved data and cookies, separate from normal browsing.
- feat: Bound how many images stay in context. Older tool screenshots can be omitted automatically, pinned images stay, and an omitted image can be attached again.
- feat: Quote selected message text into the composer, jump back to the source, and translate the selection in the app.
- feat: Copy an assistant in full, or choose memories, phrases, world books, skills, tools, and the default workspace.
- feat: Let each conversation choose its own MCP servers, or keep following the assistant.
- feat: Show the request URL while editing a provider, fill in a missing API version, and explain a failed connection test.
- feat: Remember which models from an OAuth account are enabled across syncs and sign-ins.
- feat: Encrypt new backups with Argon2id, and remember the encryption setting. Existing encrypted backups still restore.
- feat: Add ElevenLabs and Qwen speech transcription, including real-time, recorded audio, and Qwen Token Plan.
- feat: Add an optional legacy inline code style, with a lighter background, a thin border, and more line spacing.
- fix: Keep Gemini thought signatures with the provider that created them. Switching provider or model during a tool call starts a new reply instead of continuing the old round.
- fix: Continue from a context summary in a new chat without bringing back the messages that summary replaced.
- fix: Keep forked messages in their original order.
- fix: Keep a workspace file path on the message that produced it.
- fix: Match OAuth limits and usage with current Codex, Claude, Kimi, and Grok accounts, and show where sign-in failed.
- fix: Keep chats when storage is full, and recover from a local copy instead of opening an empty list.
- fix: Keep Android file saves when the activity is recreated.
- fix: Keep attachments intact while a backup is written, and keep the open chat visible when switching conversations.
- fix: Speed up long conversations with many tool calls, and keep their place while they update.
- fix: Render Mermaid diagrams in the system web view, stop block formulas from clipping, and keep Markdown tables and text readable on custom backgrounds.
- fix: Fix the in-app browser on iOS, macOS, and Windows, including toolbar insets, in-page search, clicks, and shutdown.
- fix: Read current memory in the iOS sandbox, and keep Node fetch working.
- fix: Accept MCP sign-in when OAuth discovery is nonstandard or the issuer URL has a trailing slash.
- fix: When a search provider reads full pages, point users to turn off the built-in page fetch so the tools don't overlap.
- fix: Keep IntelligentUI cards steady while they stream, and fix chart spacing.
- fix: Publish Linux builds with and without Impeller.
- fix: Fix several other known bugs and UI issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.3.2...v1.3.3

What's Changed
- feat: Rebuild Markdown rendering, with GitHub alerts, directive cards, inline HTML, themed highlights, math, and bounded SVG.
- feat: Copy a message as Markdown, and copy a selection as rich HTML.
- feat: Add an integrated browser with tabs, bookmarks, history, downloads, and agent browser tools. Browser steps appear as page cards on the timeline, with workspace page previews.
- feat: Add IntelligentUI, including layouts, forms, charts, tables, and restricted web previews, with a built-in visualize skill.
- feat: Add reversible context management. Compress a chat into an editable summary or mark a context boundary, then undo either one. Original messages are kept.
- feat: Add per-conversation tool permissions, with ask-for-approval and full access.
- feat: Add grouped MCP server management and a unified tools panel.
- feat: Add optional password encryption for backups.
- feat: Add a custom data directory on Windows.
- feat: Copy world book entries, and move them between books in a batch.
- feat: Add an optional native text selection menu on iOS 26 and later.
- feat: Add DeepSeek web search and full-page fetch, including publication dates.
- feat: Add Groq and Mistral speech transcription.
- feat: Choose memory types that are always injected in full, ignoring the per-type item limit.
- feat: Add Gemini 4 support, including thinking levels and thought-signature replay.
- fix: Replay Gemini tool calls one round at a time, keep each round's text and thinking, and keep code-execution IDs unique.
- fix: Correct Claude adaptive thinking modes and model defaults, and leave synthetic citation cards out of history.
- fix: Keep Responses tool history in order, retain tool images, and merge reasoning fragments before replay.
- fix: Send MCP tool images when a turn continues or history is replayed.
- fix: Keep assistant identities and their chats attached after an upgrade.
- fix: Refresh the Codex model list after the app version changes.
- fix: Preserve Android IME edits in long messages, including select all, cut, and composition.
- fix: Stop Windows from attaching the same image twice on a newly sent message.
- fix: Keep long-image exports at full precision, and speed up their previews.
- fix: Keep data when an import or migration fails, recover a missing installation, and resume a Windows restore blocked by a locked folder.
- fix: Leave iOS app data out of system backups.
- fix: Keep the desktop composer focused when switching windows, and restore title-bar double-click zoom on macOS.
- fix: Follow the bottom of a streaming reply directly, instead of delayed auto-scroll.
- fix: Run the sandbox on 32-bit Android, and pass the app proxy into guest processes.
- fix: Keep calculator results exact, correct GPT-5 reasoning defaults, and enable reasoning for OpenRouter Auto.
- fix: Normalize Qwen WAV audio before playback.
- fix: Sign and notarize macOS release builds.
- fix: Speed up search, long histories, avatar loading, and reasoning-slider dragging.
- fix: Fix several other known bugs and UI issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.3.1...v1.3.2

What's Changed
- feat: Add a models.dev model catalog and a unified model editor for input modalities, context window, pricing, and reasoning.
- feat: Pick reasoning effort from each model's supported levels, with a discrete slider and an optional badge on the input bar.
- feat: Show context usage on the input bar and the mobile header, with a stacked breakdown, estimated cost, and cost totals on the statistics page.
- feat: Show first-token latency and generation speed, with an option to sum token usage across a whole turn.
- feat: Send native audio, video, and PDF inputs, and record or play audio attachments.
- feat: Call embedding APIs for embedding models on OpenAI-compatible, Gemini, and Vertex providers.
- feat: Add Vercel AI Gateway, including provider-native web search and citations.
- feat: Add an optional conversation-level prompt cache key for OpenAI-compatible requests.
- feat: Add Exa MCP search, with an optional API key.
- feat: Read full web pages during search, using the search provider or a free local fetch.
- feat: Import SillyTavern world books.
- feat: Keep a separate input draft for each conversation, including attachments.
- feat: Choose which data to include when exporting or importing backups, for local, WebDAV, S3, Cherry, and Chatbox.
- feat: Add a workspace image viewing tool with inline previews in chat.
- feat: Import provider configs from the desktop providers page.
- fix: Replay complete reasoning and tool-call history by default, including Claude, DeepSeek, and the Responses API, and keep it after message edits and tool recovery.
- fix: Fix MCP OAuth discovery and browser authorization, and preserve tool schema constraints.
- fix: Fix phone control on live video, and stop the overlay capsule from stealing taps.
- fix: Keep haptic pulses going while a reply streams.
- fix: Stop the expanded composer from covering the title bar.
- fix: Bundle the runtime libraries required by the Linux AppImage.
- fix: Update Claude Code compatibility for Opus 5.5, and send the system prompt before messages.
- fix: Merge Grok OAuth model catalogs with their context limits.
- fix: Ask for confirmation before deleting legacy memories.
- fix: Improve Bing local search requests and result parsing.
- fix: Deep-merge custom request bodies so partial objects no longer overwrite generated fields.
- fix: Speed up long chat histories, ZIP backups, long-message editing, and Markdown previews.
- fix: Fix several other known bugs and UI issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.3.0...v1.3.1

What's Changed
- feat: Add workspaces with desktop terminals, Android and iOS sandboxes, skills, and mobile MCP stdio.
- feat: Add scheduled assistant tasks on Android, desktop, and iOS.
- feat: Add mobile background execution with notifications, overlays, and Live Activities.
- feat: Add ChatGPT, Grok, and Kimi Code OAuth login, with mobile browser sign-in and cache TTL settings.
- feat: Add Accessibility phone control on Android.
- feat: Add Kimi Basic, Kimi Pro, and Kagi search providers.
- feat: Add settings search.
- feat: Add sticky, cooldown, and delay effects for world books, plus per-conversation prompts.
- feat: Add reordering for TTS and speech recognition services.
- feat: Preview SVG code blocks inline.
- feat: Collapse long user messages behind an expand toggle.
- feat: Restore Google Fonts with on-demand downloads.
- feat: Add assistant gradient backgrounds.
- feat: Show the context message count in the context management sheet.
- feat: Restore backup snapshots from startup recovery.
- feat: Add an optional ISO 8601 format for appended message time.
- feat: Add an option to hide the system title bar on Linux.
- feat: Adapt DeepSeek V4.1 Flash, MiMo v2.6 vision, and Grok 4.7 reasoning.
- fix: Bundle the MSVC runtime so Windows can launch without a separate Visual C++ install.
- fix: Restore the Windows window position after display changes, and fix clipboard history paste.
- fix: Fix Gemini tool follow-ups by dropping unsigned empty text and keeping the text thought signature.
- fix: Keep hosted search cards inline, and migrate Grok built-in search to Responses tools.
- fix: Render inline thinking next to tool calls, and keep reasoning state when creating a conversation.
- fix: Honor the Qwen Audio capture sample rate, and ignore premature empty Google STOP frames.
- fix: Paginate memory_read, send custom provider headers when listing models, and stop the default SiliconFlow balance lookup.
- fix: Improve Android high refresh rate compatibility.
- fix: Honor the chat background mask strength on tablets.
- fix: Speed up streaming and long-message Markdown rendering.
- fix: Fix several other known bugs and UI issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.2.6...v1.3.0

What's Changed
- feat: Redesign the mobile reasoning budget as a stepped slider.
- feat: Add clear actions for imported fonts and local models in Storage Space.
- feat: Adapt GPT-6 Astra, Claude 5.1, Qwen 3.8, Muse 1.3, and GLM-5.3.
- feat: Add a current-location tool on Android.
- feat: Add menstrual records to health data, and expand sleep summaries with time in bed, awake periods, and sleep stages.
- feat: Make per-chat models optional and off by default; when off, picking a model applies to the current assistant.
- feat: Add a setting to send Markdown image links as images, off by default.
- fix: Send the lowest legal effort when turning thinking off on always-on models, and always send Laguna thinking knobs.
- fix: Send stable session headers to OpenCode.
- fix: Retry failed tool follow-ups before output, and reuse generated images in follow-up edits.
- fix: Fix the restore black screen on large backups, and stop replaying superseded memory snapshots.
- fix: Fix Markdown paragraph breaks and bubble sizing, table scrolling during streaming, math rendering, and TTS reading of code blocks.
- fix: Enable high refresh rates on Android, and fix several other known bugs and UI issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.2.5...v1.2.6

What's Changed
- feat: Add device-local assistant tools on iOS (weather, location, reminders, and HealthKit with per-type selection), plus a native translation selection action.
- feat: Add Claude web fetch and code execution server tools, bringing generated files into the chat and keeping the container across rounds.
- feat: Add per-conversation model overrides, and default title/suggestion models to the current chat model.
- feat: Add local database snapshots and more resilient backups, with detailed failure reports on the recovery screen.
- feat: Add AnySearch, Parallel, and You.com search providers, plus Brave LLM Context mode.
- feat: Add optional exponential-backoff auto-retry for failed API requests.
- feat: Let kelivo_fetch POST to user-supplied APIs.
- feat: Let users rewrite built-in tool schema descriptions.
- feat: Add a setting for fit-content assistant bubble width.
- feat: Add an option to split assistant replies into per-paragraph bubbles.
- feat: Add an optional surface ladder so cards sit above a sunk page.
- feat: Drop thinking/tool cards on assistant edit by default, with a setting to keep them.
- feat: Store Gemini thought signatures as provider artifacts instead of in message text.
- fix: Unlock the composer send button after switching conversations, so messages can be sent in parallel across chats.
- fix: Stop MCP Streamable HTTP GET 404 reconnect storms.
- fix: Route every Gemini 3.x model through thinkingLevel.
- fix: Echo DeepSeek reasoning on tool continuations.
- fix: Bound tool-image decode and persist MCP results as Markdown.
- fix: Improve CJK font fallback in math rendering.
- fix: Fix several other known bugs and UI issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.2.4...v1.2.5

What's Changed
- feat: Add per-role user/assistant bubble style overrides, with independent editing and live preview.
- feat: Add a setting to keep the screen on during generation.
- feat: Add settings to hide thinking cards, tool cards, and markdown images inside tool cards.
- feat: Improve the soft sheen on in-progress thinking and tool cards.
- feat: Add event reminders for calendar_create.
- feat: Adapt DeepSeek vision and thinking efforts, Gemini 3.7, Grok 4.6, and Laguna thinking/tool support.
- fix: Cut send latency and scope the file parsing indicator to the owning message.
- fix: Make title summarization opt-in unless a title model is set, so unconfigured users no longer get a failure toast after every reply.
- fix: Stop backup export from hanging, with cancellable progress and hardened Android backup streaming.
- fix: Fix frosted timeline snapshot lifecycle, grouping, and tool extent estimates.
- fix: Preserve reply position when regenerating, and keep collapsed reasoning when editing assistant text.
- fix: Decode Step SSE transcripts as UTF-8, and preserve malformed SSE text.
- fix: Fix several other known bugs and UI issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.2.3...v1.2.4

What's Changed
- feat: Add keep-recent context compression so recent turns stay verbatim, and pick the compress model from the compression dialog.
- feat: Add conversation multi-select in the drawer, with batch pin, move, and delete.
- feat: Add Chatbox ZIP v2 backup import.
- feat: Add a dedicated message style settings page with live bubble preview and color overrides.
- feat: Add configurable memory injection, humanized outcomes, editable type/scope, and a rebuilt about page, along with migration prompt and batch-size settings.
- feat: Add a switch to keep message versions when forking.
- feat: Add DeepSeek search and OpenRouter server tools.
- feat: Add export for leftover Hive chat files after SQLite migration.
- feat: Add a setting to disable or retune long-paste-as-file.
- feat: Add official GLM-OCR support with empty-prompt handling.
- feat: Raise the assistant context message limit to 4096.
- feat: Rebuild the streaming layer, matching streaming markdown to the finished render.
- feat: Remove Google Fonts support.
- fix: Fix backup file-lock contention that could lock the app out of its own data.
- fix: Fix minimap search to use full message text, with stable snippets and debounce.
- fix: Prevent MCP tool name collisions with built-in tools, and send SSE tool calls as UTF-8.
- fix: Preserve Gemini thought signatures and rebuild Claude thinking blocks from reasoning_details.
- fix: Sum token usage across tool-call rounds.
- fix: Fix several other known bugs and UI issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.2.2...v1.2.3

What's Changed
- feat: Add a global switch for legacy memory mode.
- feat: Add context logging for assembled model messages, with secret redaction and large-payload elision.
- feat: Allow Firecrawl search without an API key.
- feat: Improve streaming chat performance.
- fix: Keep the chat timeline stable when deleting mid-list or last messages.
- fix: Keep version groups anchored when deleting the anchor revision, and create a new reply after all versions are deleted.
- fix: Avoid invalid Claude thinking signatures.
- fix: Force non-streaming utility text generation.
- fix: Correct grouped provider reorder indices.
- fix: Fix several other known bugs and UI issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.2.1...v1.2.2

What's Changed
- feat: Add screen time (iOS only) and calendar local device tools.
- feat: Add Azure speech (TTS) support.
- feat: Add Doubao web search support.
- feat: Add support for duplicating topics.
- fix: Harden Hive-to-SQLite migration and backup restore against corrupt or legacy field values.
- fix: Align MiMo and DeepSeek Responses support.
- fix: Support OpenRouter Responses image generation.
- fix: Improve reasoning content replay policies.
- fix: Support Gboard image paste in chat.
- fix: Preserve scroll position when paging history, and keep the chat tail stable after deleting the latest message.
- fix: Expose app documents in the iOS Files app.
- fix: Fix several other known bugs and UI issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.2.0...v1.2.1

What's Changed
- feat: Migrate chat and business data persistence to SQLite, with hardened legacy migration and backup merge.
- feat: Rebuild the memory system — data models, tools, prompt injection, background organize pipeline, high cache hit rate, run traces, and legacy migration.
- feat: Add user custom themes with a full color picker and semantic color tokens.
- feat: Add configurable speech recognition (ASR) services and a voice input UI with live mic amplitude.
- feat: Add TTS replay and audio export.
- feat: Add MCP OAuth authorization.
- feat: Redesign the search service settings page, with multi-key rotation and Grok reasoning effort configuration.
- feat: Store message attachments as structured parts and portable, stable URIs.
- feat: Automatically attach long pasted text as files.
- feat: Add image compression for uploads and persist image OCR cache.
- feat: Add provider custom request configuration.
- feat: Allow configuring thinking/reasoning for utility models (title generation, OCR, etc.).
- feat: Add recovery actions on the fail-closed startup screen (repair, export data copy, reset).
- feat: Enable process restart on desktop.
- feat: Add image filtering and sorting in storage management.
- feat: Improve mobile background layout and iPad floating window support.
- feat: Adapt latest model capabilities and expand search/voice providers (StepFun, Firecrawl, TinyFish, Step/Fish/Qwen Audio, etc.).
- fix: Fix stuck streaming, regeneration, version switching, and temporary conversation persistence races.
- fix: Fix MCP remote sessions, duplicate tool names, and legacy SSE compatibility.
- fix: Fix stream cancellation responsiveness and support additional thinking tags.
- fix: Fix several other known bugs and UI issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.17...v1.2.0

What's Changed
- feat: Add support for Claude Opus 4.8, Fable 5, Kimi K2.7, GLM 5.2, and Agnes image models.
- feat: Add "Maximum Reasoning" budget level for Claude 4.7+ and Fable models.
- feat: Add "Enable Thinking" toggle for title generation.
- feat: Add reasoning/thought extraction support for Google Gemini models.
- feat: Add provider avatar icon selection with Metaso icon support.
- feat: Add customizable chat input background opacity.
- feat: Add option to batch delete models that failed detection.
- feat: Add mobile message navigation buttons with three display modes.
- feat: Add User-Agent header support for WebDAV and S3 backup settings.
- feat: Add calculator local tool for MCP.
- feat: Add cross-platform stdio command resolution and Windows batch support for MCP.
- feat: Add initial model selection support in model selector.
- fix: Fix markdown parser misinterpreting HTML tags inside fenced code blocks.
- fix: Fix TeX rendering escaping # in hex color arguments.
- fix: Fix desktop text selection lost on scroll.
- fix: Fix loaded chat window not containing visible messages.
- fix: Fix UI display issue when no session is present.
- fix: Improve reasoning content handling and drag-to-dismiss gesture physics.
- fix: Fix several other known bugs.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.16...v1.1.17

What's Changed
- feat: Improve user message editing, including support for uploading attachments while editing.
- feat: Add message multi-selection and batch deletion.
- feat: Add provider model multi-selection and batch deletion.
- feat: Improve citation source popup.
- feat: Improve Text-to-Speech features, including floating player, local TTS tool, auto-play, text selection settings, and Qwen, Groq, and xAI network TTS services.
- feat: Add iOS background generation with Live Activities support.
- feat: Add DeepSeek built-in search support in Claude format and DeepSeek Claude-compatible reasoning support.
- feat: Add Querit search provider support.
- feat: Add Cherry Studio v6 direct backup import support.
- feat: Add low-memory streaming ZIP backup and S3 download handling.
- feat: Add configurable TTL for Claude prompt caching.
- feat: Add HTTP 307/308 redirect support in mcp_client.
- feat: Add per-assistant web search toggle settings.
- feat: Improve Markdown rendering, including inline LaTeX, CJK compatibility, table delimiter handling, code block controls, and PlantUML block styling.
- feat: Improve image viewer and image preview.
- feat: Improve model selector with sticky provider headers and synchronized tab scrolling.
- fix: Fix UTF-8 decoding for non-streaming API responses.
- fix: Fix pending tool events incorrectly creating API tool calls.
- fix: Fix token usage parsing and add support for caching metrics.
- fix: Fix Claude model override and multi-block response handling.
- fix: Fix image input filtering for text-only models and OCR model capability validation.
- fix: Fix image parsing incorrectly handling code blocks and inline code.
- fix: Fix linked conversation cleanup when deleting assistants.
- fix: Fix several other known bugs and UI issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.15...v1.1.16

What's Changed
- feat: Add support for Gemini 3.5 Flash.
- feat: Improve Mermaid diagram rendering stability.
- fix: Fix Markdown inline math rendering issues.
- fix: Fix missing function/tool call blocks during Claude and OpenAI tool-call continuation.
- fix: Fix date drift in stats heatmap.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.13...v1.1.15

What's Changed
- feat: Add OpenAI Images API generation and editing support, with toggleable image mode for OpenAI and SenseNova image models.
- feat: Add optional image cropping after image selection.
- feat: Add Serper and Grok search provider support.
- feat: Add built-in search support for OpenRouter.
- feat: Add reasoning/thinking support for GPT-5.5, DeepSeek, and Google Gemma 4.
- feat: Add Claude Prompt Caching support for Anthropic and OpenRouter.
- feat: Add provider-specific tool metadata support for Claude and Gemini.
- feat: Add chat statistics page with usage trends, heatmap, rankings, and token metrics.
- feat: Add account balance query support for OpenAI-compatible providers.
- feat: Add configurable context compression options.
- feat: Add regeneration behavior settings, including trailing-message deletion and confirmation controls.
- feat: Add periodic backup reminder system.
- feat: Add chat follow-up suggestions.
- feat: Add assistant local tools, including clipboard, time info, and interactive ask_user questions.
- feat: Add MCP configuration tab to assistant settings.
- feat: Add customizable assistant edit tab ordering and section-list outline mode.
- feat: Add temporary chat mode.
- feat: Add lazy loading, sliding message window, and bidirectional loading for long chat histories, improving performance with very large message counts.
- feat: Add smooth text animation for streaming responses.
- feat: Enhance Markdown rendering with support for HTML details/summary/link tags, nested details, improved code block experience, and table export actions.
- feat: Add configurable visibility modes for desktop message navigation buttons.
- feat: Add desktop setting to show provider names in chat messages.
- feat: Set Linux GTK application and window icons.
- fix: Fix macOS window restoration and activation behavior.
- fix: Fix desktop hotkeys being included in data synchronization.
- fix: Fix Gemini thought signatures and internal part IDs in request payloads.
- fix: Fix image content being lost in tool-call continuation requests.
- fix: Fix selected-message export not using stored messages.
- fix: Fix several other known issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.12...v1.1.13

What's Changed
- feat: Add support for DeepSeek V4 model and update reasoning parameters.
- feat: Add support for Claude 4.7 model and dynamic web search.
- feat: Add built-in search support for Qwen 3.6 models on DashScope.
- feat: Support reordering the "Other" provider group and improve provider header drag-and-drop.
- fix: Fix COS failing to list restore entries.
- fix: Fix window display issue on multi-monitor setups on Windows.
- fix: Prevent MCP tool failure from terminating chat flow.
- fix: Exclude chat font scale from data synchronization.
- fix: Eliminate IM candidate window offset on KDE.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.11...v1.1.12

What's Changed
- feat: Add manual approval mode for MCP tool execution, along with updated approval settings UI.
- feat: Add timeline-based rendering for reasoning and tool calls, with support for <thought> tags and dynamic reasoning budget icons.
- feat: Add delete-all-versions action for multi-version messages.
- feat: Add detailed token statistics popup with enhanced tracking, adaptive positioning, and animations.
- feat: Add separate display toggles for user/model names and timestamps.
- feat: Add MiMo TTS service support.
- feat: Add X-Conversation-Id header support for API requests.
- fix: Fix stuck streaming and reasoning states, with improved crash recovery and pending-send handling.
- fix: Fix Claude thinking requests by correctly handling missing budget_tokens in translation and OCR flows.
- fix: Preserve trailing messages when regenerating from earlier messages.
- fix: Fix desktop window restore position after display changes.
- fix: Fix several other known bugs and UI issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.10...v1.1.11

What's Changed
- fix: Fix edited messages not updating immediately after save.
- fix: Fix reply switching showing the wrong regenerated content.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.9...v1.1.10

What's Changed
- feat: Add context compression feature with customizable model and prompt.
- feat: Add global search mode for conversations.
- feat: Add S3 backup support.
- feat: Add support for DashScope built-in search and reasoning tools.
- feat: Add custom API URL support for Exa and Tavily search services.
- feat: Add support for Google Vertex AI Claude models and Claude 4.6 adaptive thinking.
- feat: Add support for GPT-5 series models and reasoning levels.
- feat: Add support for Moonshot Kimi thinking models and kimi-k2.5 reasoning echo.
- feat: Add support for multimodal LongCat Omni models.
- feat: Add reordering support for World Books and entries.
- feat: Add setting to toggle Markdown rendering for assistant messages.
- feat: Add "Show Provider" option to chat messages.
- feat: Add assistant name and avatar display options for chat titles.
- feat: Add confirmation dialogs for message regeneration and topic deletion.
- feat: Add log management settings and auto-cleanup.
- feat: Add customizable summary refresh frequency for assistants.
- feat: Increase maximum context message size to 1024.
- feat: Add scroll-to-bottom button in mini map.
- refactor: Improve chat auto-scroll.
- refactor: Improve citation display with card-style search results.
- perf: Optimize frosted chat bubble rendering performance.
- perf: Optimize backup and restore memory usage.
- fix: Fix several other known bugs.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.8...v1.1.9

What's Changed
- feat: Added proxy bypass setting.
- feat: Added delete confirmation for remote backups.
- feat: Added Ctrl/Cmd+Enter shortcut for newlines.
- feat: Support <br> tag rendering.
- feat: Support multi-API key import for Cherry Studio.
- feat: Added structured request log viewer.
- feat: Enable text selection in reasoning body.
- feat: Added MCP base64 image handling and tool card image display.
- fix: Validate STDIO command existence before launch to prevent crash on macOS.
- fix: Fixed model detection failure when duplicate model IDs exist.
- fix: Fixed Responses API tool call history conversion.
- fix: Fixed iPad window control overlap issue.
- fix: Fixed dollar signs in inline code being misparsed as LaTeX.
- fix: Fixed message send failure after tool usage.
- perf: Improved conversation export page.
- perf: Cached extracted document text and optimized parsing strategy.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.6...v1.1.7

What's Changed
- feat: Added a toggle switch for logging.
- feat: Model IDs can now contain spaces.
- feat: Added a storage management page.
- feat: Added built-in tools that support the official API.
- feat: Added support for OpenRouter's thinking signature.
- feat: Allow listing models without requiring an API key.
- feat: Enhanced the chat history reference feature, now includes LLM-powered conversation summaries.
- feat: Allow scrolling by clicking and dragging in the blank space of the chat area on desktop.
- feat: Added an "Enter to Send" option for mobile devices.
- feat: Added configurable sending shortcuts for the desktop version.
- fix: Fixed several other known bugs.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.5...v1.1.6

What's Changed
- feat: Add a setting to toggle auto-scrolling.
- feat: Support tool calls for DeepSeek & Claude reasoning models.
- feat: Add a reset button for the assistant chat model.
- feat: Add timeout configuration for MCP tool calls.
- feat: Add desktop shortcuts for switching models and closing windows.
- feat: Support multi-line replacement in regex settings.
- feat: Add scroll navigation buttons.
- feat: Add a setting to automatically collapse code blocks.
- feat: Adjust the minimum chat font size limit to 50%.
- fix: Fix stuttering issues after message generation completes.
- fix: Fix flickering in the dialog menu.
- fix: Fix various known issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.3...v1.1.5

What's Changed
- feat: Add batch model availability testing.
- feat: Add support for Regular Expression (Regex) replacements.
- feat: Allow importing system prompts and instruction injections from files.
- feat: Allow manually specifying context message limits.
- feat: Add an expand/collapse button to the chat input field.
- feat: Remember the last used translation language.
- feat: Add "Save & Send" capability to message editing.
- feat: Add option to delete all models for a specific provider.
- feat: Support duplicating/copying assistants.
- refactor: Update the message generation animation.
- refactor: Optimize Markdown rendering.
- refactor: Improve various operational interactions and logic.
- fix: Fix general known issues.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.2...v1.1.3

What's Changed
- feat: Allow selecting streaming mode when testing connections.
- feat: Add support for gemini-3-pro-preview and gemini-3-pro-image-preview models.
- feat: Add support for PlantUML rendering.(https://github.com/Chevey339/kelivo/pull/102)
- feat: Add DuckDuckGo as a search provider (no API key required).
- feat: Add support for built-in search capabilities in Grok models.
- refactor: Instruction Injection is now configured individually per assistant.
- refactor: Optimize the logic for showing and hiding the application on Windows.(https://github.com/Chevey339/kelivo/pull/103)
- fix: Fix various known bugs.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.1...v1.1.2

What's Changed
feat: Add keyboard shortcuts for the desktop client.
feat: Introduce "Instruction Injection" feature, which replaces the previous "Learning Mode".
feat: Add a system tray menu for quick access (desktop).
feat: Allow resetting the default model to its original setting.
feat: Allow setting custom avatars for providers.
feat: Add the ability to move conversations between different assistants.
feat: Add support for uploading video files (for models with video input capabilities).
feat: Introduce Image OCR (Optical Character Recognition) functionality.
feat: Add a setting for automatic word wrap in code blocks (mobile only).
feat: Add an assistant_name variable available in the system prompt.
fix: Resolve an issue where non-streaming chat mode could not be enabled correctly.
fix: Addressed several other known bugs to improve stability.
perf: Implemented lazy loading for system fonts to reduce memory consumption.
refactor: Optimized code block functionality and user experience.
refactor: Polished and improved several UI pages.
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.1.0...v1.1.1

What's Changed
- feat: Add support for desktop platforms.
- feat: Introduce a built-in MCP tool (fetch).
- feat: Add preset information for assistants.
- feat: Integrate Text-to-Speech (TTS) services from OpenAI, Gemini, and ElevenLabs.
- feat: Add a "Save" button to the image preview screen.
- feat: Introduce a global proxy setting.
- feat: Allow the default assistant to be deleted.
- feat: Add a new translation page with a Spanish language option.
- feat: Add the ability to set custom fonts.
- feat: Add a new setting for chat message backgrounds.
- feat: MCP now supports JSON editing.
- feat: Add a feature to group assistants.
- feat: Enable background conversation generation on Android.
- feat: Add HTML preview functionality.
- feat: Add a web view preview for messages.
- fix: Addressed several known bugs
Full Changelog: https://github.com/Chevey339/kelivo/compare/v1.0.16...v1.1.0