Chatbox: baixar versions anteriors

1. Fixed external one-click MCP setup links failing to open the prefilled server dialog when the app was not running

1. When a reply is cut off, you'll see why and how to continue
2. Fixed thinking controls for DeepSeek V4.1 Flash
3. Added Turkish as an interface language
4. Image generation model pickers now show model icons
5. Improved OpenCode Go and Zen connectivity on mobile and the web

1. Added OAuth authorization for remote MCP servers, with connection controls and a refreshable tool list in server settings
2. Added custom conversation compaction prompts and one-time instructions, with improved compaction scope and trigger accuracy
3. Added cached token usage to message statistics and configurable prompt cache duration for Claude
4. Added support for more plain-text uploads, including Typst, Patch, and Diff files
5. Added one-click undo for archived chats and improved recovery when a conversation fails to load at startup
6. Fixed replies sometimes remaining in the generating state after completion
7. Improved generated image loading, with an option to reload images that fail to display
8. Added reasoning effort controls for GPT-6 and later models, and improved Kimi, OpenCode, and model catalog compatibility
9. Fixed backup exports on HarmonyOS and improved window focus, small-screen layouts, and translations
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. Added persistent memory: new chats can remember key facts you saved, as Global Memory or per Copilot
2. Added Soul in Work Mode to set the AI's personality and boundaries
3. You can queue a new message while a reply is generating, or insert one immediately
4. Working directory and permissions carry over from last time, and can be changed above the input
5. Full Access no longer pauses for confirmation every 25 steps by default
6. Vision-capable models can open local images or newly downloaded files on their own
7. Improved context for long tasks: tool results are not dropped too early, and recent messages are kept first when compacting
8. You can clear all archived chats at once; text in tool results can be previewed
9. Added Tencent Hunyuan, Xiaomi MiMo, LongCat, GLM Coding Plan, OpenCode Zen, and OpenCode Go, plus self-hosted SearXNG search
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. Fixed DeepSeek V4 Flash Vision Exp being treated as a text-only model when cached model metadata is outdated
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. Fixed a performance issue that could cause lag when regenerating replies or switching response branches in long conversations
2. Improved backup export compatibility on devices where backups could fail to export
- MacOS(Apple Silicon) beta.1
- MacOS(Intel) beta.1
- Windows beta.1
- Android beta.1
- Linux x86_64 beta.1
- Linux ARM64 beta.1

1. Refined the classic message layout and overall interface, with customizable accent colors and color presets
2. Organized alternative replies into collapsible branches for easier comparison and clearer conversation context
3. Improved Work Mode for long-running tasks, approvals, and command compatibility
4. Added Claude Opus 5 support and improved DeepSeek thinking compatibility
5. Fixed issues with file parsing, automatic scrolling, Mermaid rendering, and overall stability
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. Added Work Mode: the AI can run code, use tools, work with files on your computer, find past conversations, and generate images to complete complex tasks. Learn more
2. Added Skills: install and manage reusable AI abilities, with a regularly updated built-in library
3. Added complete data backups that include conversations, images, and attachments for export and restore
4. Redesigned the model picker with search and per-model thinking depth
5. Added a conversation minimap for navigating long chats
6. Refreshed onboarding with ready-to-use examples
7. Added chat archiving and creation time in the sidebar
8. Added PDF support on mobile
9. Improved streaming responses, automatic scrolling, and reasoning status display
10. Updated built-in DeepSeek models and improved compatibility with MCP servers
11. Various bug fixes and stability improvements
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. Fixed an issue where the send button could stay disabled after uploading files
2. Updated the Android app icon
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. Added large-file attachment Q&A with indexing and on-demand retrieval for text files.
2. Improved attachment upload, indexing status, retry flow, and file parsing errors.
3. Upgraded image generation with dynamic image model lists and clearer error messages.
4. Added Vercel AI Gateway provider and improved provider search and selection.
5. Improved web search and translation reliability, especially for mainland China networks.
6. Redesigned auto-update experience with clearer progress, retry, and install flows.
7. Improved session list performance and stability for large chat histories.
8. Fixed issues with DeepSeek V4 thinking mode, attachments, session deletion, and code block copying.
- MacOS(Apple Silicon) beta.4
- MacOS(Intel) beta.4
- Windows beta.4
- Android beta.4
- Linux x86_64 beta.4
- Linux ARM64 beta.4

1. Added verification code login support
2. Added AWS Bedrock provider support
3. Added BoCha and Querit.ai web search providers
4. Added Tavily parse_link tool for richer web search results
5. Added translate button for raw error messages
6. Upgraded Mermaid with Venn diagram support
7. Adjusted input box font size
8. Fixed conversation history losing tool-call parts
9. Fixed input content lost when max context message count is 0
10. Fixed title bar disappearing when all chats are cleared
11. Fixed Gemini 3 models thinking mode configuration
12. Fixed macOS input focus causing empty bottom margin
13. Fixed Android large file export auto-rename
14. Removed deprecated Gemini OAuth support
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. New chat bubble mode for a more natural conversation experience
2. Unified model list that automatically fetches the latest model information from all providers
3. Redesigned thinking process and web search display for clearer AI reasoning visualization
4. New Copilot system with pre-configured AI assistants for common tasks
5. Switched to Shiki for code syntax highlighting with better language support and themes
6. Support session background images for personalized chat experience
7. Added OAuth login for Qwen, MiniMax, Moonshot, and Anthropic providers
8. Added setup guidance for message layout settings
9. Improved sidebar layout on small screens
10. Font size setting now only affects chat messages instead of the entire UI
11. Fixed token usage display now showing accurate counts from AI provider
12. Fixed image generation stability on Android
13. Multiple bug fixes and UI improvements
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. Image generation tool now supports Nano Banana 2
2. Fixed drag-and-drop file parsing on desktop
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. New context management: automatic context compression, token usage percentage display, and context length error detection
2. Default max context messages changed to unlimited
3. AI SDK upgraded from v5 to v6
4. Image generation refactored into a standalone tool page
5. Chatbox AI settings moved from provider level to top-level settings for easier access
6. Brand new conversational onboarding experience
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. Fixed compatibility issues on some older devices
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. Added model icon display in model selector
2. Added search functionality for conversation list
3. Auto-enable web search when using Chatbox AI
4. Knowledge base file parsing now supports backend parsing as fallback
5. Auto-focus message input when window gains focus
6. Improved mobile UI experience in multiple areas
7. Added visual indicator for expired License status
8. Added automatic retry mechanism for API 5xx errors
9. Fixed code block visibility issue in light mode
10. Fixed global search navigation to target location
11. Fixed adaptive modal height issue
12. Fixed message edit modal close issue
13. Fixed an issue where assistant messages would sometimes get a > prefix added
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. Added support for tool calling with the new deepseek-reasoner model
2. When file parsing fails, you can click on the file to view the error reason
3. Prevented selection of unsupported file types when uploading files
4. Fixed an issue where messages could be sent even when file parsing was incomplete or failed
5. You can now view the parsed content of uploaded files in conversations
6. Added diagnostic log export feature in settings, making it easier for users to send logs for investigating hard-to-reproduce issues
7. Fixed an issue where conversation name and system prompt couldn't be set when the sidebar was open on mobile
8. Fixed an issue where Gemini provider was not sending temperature and topP parameters
9. Fixed image generation support for gemini-3-pro-image-preview and gemini-2.5-flash-image in Gemini provider
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. Added email login support, allowing users to use licenses bound to their email after logging in
2. Added support for OpenAI Responses API
3. Optimized code styling and generated webpage preview effects
4. Generated webpages can now be published with a one-click share link
5. Fixed an issue where deleting a new thread would also delete its branches
6. Custom models can now automatically detect vision and tool calling capabilities
7. Fixed a crash issue when importing data with too much historical data
8. Desktop sidebar width is now adjustable
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. Optimized storage performance, reducing lag and crashes during message generation
2. Fixed the issue where duplicate token estimation calculations caused lag
3. Fixed the issue where message format errors in certain cases caused the app interface to crash
4. Added Gemini and Claude-style APIs to custom providers
5. Optimized the styles of sidebar, conversation page title bar, and message action area, unified default avatars
6. Fixed the issue where some icons did not scale with the global font size
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. Fixed issue where the desktop message navigation component blocked the scrollbar
2. Optimized the message navigation component on mobile
3. Fixed issue where setting prompt and avatar when creating a new conversation didn't take effect
4. Fixed issue where generation couldn't be stopped when messages were generated in the middle of the list
5. Fixed issue where Artifact preview couldn't go full screen on some devices
6. Fixed issue where certain data exceptions caused the app to get stuck on the error page; you can now click the Reload App button to return to the home page
- MacOS(Apple Silicon)
- MacOS(Intel)
- Windows
- Android
- Linux x86_64
- Linux ARM64

1. Optimized model selector style with support for collapsible provider grouping
2. Added estimated input token display in the input box
3. Added conversation compression feature that can compress conversation context through summarization to reduce token consumption in subsequent conversations
4. Adjusted input box action buttons: merged image, file, and link buttons on desktop; moved create new topic to secondary menu on mobile
5. Models now support displaying and setting contextWindow, maxOutput and other properties
6. ChatboxAI image generation now supports uploading reference images
7. Fixed issue where knowledge base embedding didn't support model IDs with colon characters
8. Added Gemini provider model gemini-2.5-flash-image-preview with image generation support
9. Fixed text overflow issue with extremely long words
10. Fixed issue where mobile couldn't view complete formulas
11. Hidden action area during message generation for smoother output
12. Optimized settings page styling
13. Optimized image preview component
14. Added OpenRouter provider, supporting image generation with the 2.5-flash-image model
15. Fixed an issue where old conversation data could not be searched
16. Added message navigation buttons in the conversation window for quick jumping to previous/next messages
17. Fixed issue where modifying conversation settings or switching branches during generation would cause message status errors
⚠️ This version includes data format updates. After upgrading, conversation data generated by the new version will not be viewable in older versions (however, old version data will be preserved in the new version, and exported data from settings can be used across versions). Please upgrade with caution.

1. Fix mobile "Improve Network Compatibility" feature, support cross-domain API requests
2. OpenAI provider added support for gpt-5 series models
3. Mobile full-screen preview artifacts
4. Fix issue where input box draft was not cleared immediately after sending a message
5. Sidebar support on iOS
v1.15.3 - 2025.08.07
1. Optimize streaming output display
2. Fix message edit box width
3. Fix input box keyboard pop-up trigger condition
4. Fix message count limit reset issue
5. Optimize App system compatibility, now supports lower system versions
6. Shorten the display time of the copied prompt
7. Model list supports search
8. Input box automatically saves draft
9. Add mistral provider
10. Show thinking time
11. New conversation defaults to hiding my partner list, can be enabled in my partner page
v1.15.2 - 2025.07.24
1. Fixed issue where maxTokens parameter was missing in some clients, causing message sending to fail
2. Fixed issue where thinking content couldn't be edited
3. Fixed issue where third-party embedding API call failed in knowledge base
4. Fixed issue where some API providers returned empty thinking content, causing multiple thinking content to appear
v1.15.1 - 2025.07.22
1. Support disabling streaming output
2. Support setting max token parameter
3. Fix issue where mobile couldn't access ollama
4. Adjust input box style
5. Support importing MCP and provider configurations through deep link
6. Fix issue where new message was not sent in certain conditions

1. Optimize streaming output display
2. Fix message edit box width
3. Fix input box keyboard pop-up trigger condition
4. Fix message count limit reset issue
5. Optimize App system compatibility, now supports lower system versions
6. Shorten the display time of the copied prompt
7. Model list supports search
8. Input box automatically saves draft
9. Add mistral provider
10. Show thinking time
11. New conversation defaults to hiding my partner list, can be enabled in my partner page
v1.15.2 - 2025.07.24
1. Fixed issue where maxTokens parameter was missing in some clients, causing message sending to fail
2. Fixed issue where thinking content couldn't be edited
3. Fixed issue where third-party embedding API call failed in knowledge base
4. Fixed issue where some API providers returned empty thinking content, causing multiple thinking content to appear
v1.15.1 - 2025.07.22
1. Support disabling streaming output
2. Support setting max token parameter
3. Fix issue where mobile couldn't access ollama
4. Adjust input box style
5. Support importing MCP and provider configurations through deep link
6. Fix issue where new message was not sent in certain conditions

1. Fixed issue where maxTokens parameter was missing in some clients, causing message sending to fail
2. Fixed issue where thinking content couldn't be edited
3. Fixed issue where third-party embedding API call failed in knowledge base
4. Fixed issue where some API providers returned empty thinking content, causing multiple thinking content to appear
v1.15.1 - 2025.07.22
1. Support disabling streaming output
2. Support setting max token parameter
3. Fix issue where mobile couldn't access ollama
4. Adjust input box style
5. Support importing MCP and provider configurations through deep link
6. Fix issue where new message was not sent in certain conditions

1. Support disabling streaming output
2. Support setting max token parameter
3. Fix issue where mobile couldn't access ollama
4. Adjust input box style
5. Support importing MCP and provider configurations through deep link
6. Fix issue where new message was not sent in certain conditions

1. Local knowledge base support
2. Adjust thinking and tool call message style

1. Support MCP feature
2. Support thinking parameters for some OpenAI, Claude, Gemini models

1. Fixed storage performance issue
2. Fixed issue where clearing conversation list in English language couldn't fill in the number of conversations to keep

1. Fixed issue where custom provider couldn't set API Path
2. OpenAI, Claude, Gemini models support setting thinking effort parameter

1. Fixed image recognition issues in o1
2. Updated Perplexity model list with chain-of-thought support
3. Fixed various display issues

1. Improved the chain-of-thought display for the DeepSeek R1 model in SiliconFlow.
2. Enhanced the chain-of-thought display for the DeepSeek R1 model in LM Studio.
3. Increased the maximum number of files attachable in messages to 10.
4. Added support for o1 and o3 series models deployed on Azure.
5. Introduced support for the new o3-mini model.

1. Improved the chain-of-thought display for the DeepSeek R1 model in SiliconFlow.
2. Enhanced the chain-of-thought display for the DeepSeek R1 model in LM Studio.
3. Increased the maximum number of files attachable in messages to 10.
4. Added support for o1 and o3 series models deployed on Azure.
5. Introduced support for the new o3-mini model.

1. Fixed some minor issues.

1. Fixed some minor issues.

1. Added local document analysis. Now you can send files like PDFs, DOCs, PPTs, and XLSXs to any model API.
2. Added the option to collapse the thought process for the DeepSeek-R1 model when deployed via Ollama.
3. Fixed an issue with line breaks in the thought process display.
4. Fixed issues caused by shortcut key settings.
5. Fixed other minor issues.

1. Added support for the DeepSeek R1 model.
2. Enabled display of the model's thought process (if supported by the model).
3. Fixed the issue where Artifact previews failed due to network problems.

1. Fixed an issue with DALL-E API requests.

1. Fixed an issue on Windows where multiple application instances could be launched, causing duplicate icons in the system tray
2. On macOS, System Events permission is now only requested when enabling the "launch at startup" option, rather than automatically requesting it when the application starts
3. Fixed an issue where automatically generated titles were occasionally truncated

1. Added system tray icon, optimized show/hide shortcut
2. Input box now supports direct drag-and-drop insertion of images or files
3. Added option for launch at system startup (disabled by default, can be enabled in settings)
4. Added thread title menu for quick thread switching and deletion
5. Added toggle for automatic title generation to save tokens when disabled
6. Added Italian language support
7. Gemini and Groq models now support fetching the latest model lists remotely
8. Gemini models now support sending images by default
9. Ollama models now support sending images by default
10. Updated model lists for various model providers
11. Improved mobile experience, fixed issue with excessive bottom space after keyboard pop-up

1. Added support for new o1 model series
2. SVG image previews are now available
3. Fixed display bugs with Artifact previews
4. Fixed various minor bugs

1. Added support for setting user and assistant avatars.
2. Fixed some API compatibility issues with custom model providers.
3. Resolved various known bugs.

1. Add support for new model Claude 3.5 sonnet
2. Fixed some known issues

1. Fixed a bug where text copied from Microsoft Word was inserted as an image.

1. You can now send files to AI, with support for PDF, DOC, PPT, XLS, TXT, and code files.
2. The input box now allows you to insert images and files from the system clipboard.
3. Added support for auto-generating thread titles.
4. Now supporting the latest gpt-4-turbo model.
5. You can change the installation path during setup in Windows.
6. You can switch Gemini model version, with support for the new Gemini 1.5 pro model.
7. Fixed an issue with an abnormal message sequence in Claude causing API errors.
8. Made some UI detail adjustments.

1. You can now set user avatars in messages.
2. Added support for configuring a custom API host for Gemini.
3. Implemented an option in settings to enable or disable Markdown and LaTeX rendering.
4. Fixed issues with LaTeX rendering.
5. Fixed potential stuttering and crashing issues during message generation.
6. Fixed issues with redundant pop-up prompts during auto-updating.
7. Fixed various minor bugs.

1. Introduced support for the claude-3-sonnet-20240229 model.
2. Fixed a bug that caused blank text blocks to appear in claude model requests.
3. Fixed the auto-scroll bug in the message list that occurred when editing past messages.

1. Introducing Vision feature: You can now send images to AI!
2. Support for the latest Claude 3 series models.
3. A complete redesign of the app layout for a fresh and intuitive user experience.
4. Added message timestamps
5. Fixed some known minor issues

1. Thread Archiving (refreshes context) and Thread History List.
2. Introduced support for the Google Gemini model.
3. Introduced support for the Ollama, enabling easy access to locally deployed models such as llama2, mistral, mixtral, codellama, vicuna, yi, and solar.
4. Fixed an issue where the fullscreen window would not restore to fullscreen on the second launch.

1. Redesigned the message editing dialog
2. Fixed an issue where token configurations could not be saved
3. Fixed the positioning issue with newly copied conversations
4. Simplified the tips in settings
5. Optimized some interaction issues
6. Fixed several other issues

1. Added an image generation feature (Image Creator); you can now generate images within Chatbox, powered by the Dall-E-3 model.
2. Improved some usability issues.

1. Added direct support for the gpt-3.5-turbo-1106 and gpt-4-1106-preview models.
2. Updated the method for calculating message tokens to be more accurate.
3. Introduced the Top P parameter option.
4. The temperature parameter now supports two decimal places.
5. The software now retains the last conversation upon startup.

1. Optimized the interaction experience of the search box
2. Fixed the scrolling issue with new messages
3. Fixed some network-related issues

1. Fixed an issue where message content cannot be selected during the generation process
2. Improved the performance of the search function, making it faster and more accurate
3. Adjusted the layout style of messages
4. Fixed some other minor issues

1. Now you can search messages from the current chat or all chats
2. Data backup and restore (data import/export)
3. Fixed some minor issues

1. Keep the previous window size and position upon startup
2. Hide the system menu bar (for Windows, Linux)
3. Fixed an issue with session-specific settings causing license and other settings abnormalities
4. Adjusted some UI details

1. Automatically move cursor to the bottom of the input box when quoting a message.
2. Fixed the issue of resetting context length setting when switching models (#956).
3. Automatically compatible with various Azure Endpoint configurations (#952).

1. Automatically move cursor to the bottom of the input box when quoting a message.
2. Fixed the issue of resetting context length setting when switching models (#956).
3. Automatically compatible with various Azure Endpoint configurations (#952).

1. Support OpenAI custom models (#28)
2. The up arrow key can quickly input the previously sent message.
3. Added x64 and arm64 architecture versions to Windows and Linux installation packages.
4. Fixed issue in session settings of Azure OpenAI where the model deployment name could not be modified (#927).
5. Fixed issue with inability to enter spaces and line breaks when modifying default prompt (#942).
6. Fixed scrolling issue after editing long messages.
7. Fixed various other minor issues.

- Action buttons on messages now remain visible during list scrolling
- Added support for the Claude series models (beta)
- Language support expanded to include more countries
- Fixed some minor issues

Fixed some minor issues.

This update includes the following:
- Added application shortcuts for quickly showing/hiding windows, switching conversations, etc. See the settings page for details.
- Introduced a new setting for the maximum amount of context messages, allowing more flexible control of the context message count and saving token usage.
- Added support for OpenAI 0301 and 0314 model series.
- Added a temperature setting in the conversation settings.
- Fixed some minor issues.
If you're updating from v0.5.x to the new version, you'll need to manually uninstall the old version after installing the update.

This update includes the following:
- Added support for modifying model settings for each conversation (this allows different sessions to use different models)
- Optimized performance when handling large amounts of data
- Made the UI more compact
- Fixed several minor issues
If you're updating from v0.5.x to the new version, you'll need to manually uninstall the old version after installing the update.

This update includes the following:
- Added bulk cleaning feature for conversation lists
- Support for displaying token usage of messages
- Support for modifying the default prompt for new conversations
- Support for setting smaller font sizes
- Fixed a few other minor issues
If you're updating from v0.5.x to the new version, you'll need to manually uninstall the old version after installing the update.