
Download LobsterAI
Available for macOS and Windows.
Choose the version for your computer
- Added an optional built-in Agent browser, allowing you to view pages live on the right side of a task and take over web actions directly
- Added support for securely saving website login information and authorizing Agent auto-login when needed, with passwords encrypted by system secure storage
- Improved the app update experience with background downloads and confirmation prompts before installation may interrupt active tasks
- Improved subscription recovery for sharing and website publishing, automatically syncing resource status after subscription and providing clear guidance when redeployment is needed
- Fixed several other issues to improve application stability
Choose the version for your computer
- Added onboarding and a welcome task to help new users get started quickly.
- Improved the My Files loading experience for smoother search and browsing.
- Improved image and PPT thumbnail stability to reduce blank or incorrect previews.
- Improved voice input quota reminders to make remaining time clearer.
- Improved Windows installation and upgrade stability to reduce stalls and failures.
- Improved bot card layout in messaging channels for clearer viewing and management.
- Fixed several issues to improve app stability.
Choose the version for your computer
- Improved sharing and website management, with support for permanently deleting cloud-shared files that are no longer accessible
- Added website traffic rankings, making page views and visitor counts easy to see at a glance
- Made account benefits easier to access, with improved entry points for daily bonus credits, plan upgrades, and credit top-ups
- Simplified model selection, with older models now grouped under “More Models”
- Made update installation more reliable, so downloaded updates are no longer invalidated by temporary network disruptions
- Enhanced Windows installer validation and disk space handling, reducing the likelihood of incomplete installations and startup failures
- Improved the timeliness of sidebar activity updates, along with better icon display in dark mode and more stable model initialization
- Fixed several issues to improve overall application stability
Choose the version for your computer
- Fixed several known issues to improve system stability
Choose the version for your computer
- Added “My Files” for unified viewing and management of local artifacts, cloud-shared files, and deployed sites
- Improved the Settings page, credits summary, and plugin installation dialog experience
- Fixed an issue where Artifact previews confused web pages with local services
- Fixed several known issues to improve system stability
Choose the version for your computer
- Upgraded DeepSeek Harness to 0.1.1-rc.1
- Improved the Windows experience by fixing issues such as console windows flashing during task execution
- Fixed several bugs to improve overall application stability
Choose the version for your computer
- Fixed compatibility issues with DeepSeek Harness on certain Windows systems.
- Fixed several bugs to improve system stability
Choose the version for your computer
- Added the experimental DeepSeek Harness feature: Enable it under Settings → Experimental Features to automatically sync LobsterAI model configurations and handle complex programming tasks in a dedicated workspace
- Added intelligent task collaboration: LobsterAI can now delegate suitable multi-step programming tasks to DeepSeek Harness and return the results
- Improved subscription model loading reliability: Models now reload automatically after temporary network or login issues and when connectivity is restored, reducing instances of an empty model list
- Improved scheduled task history stability: Optimized the loading of large numbers of run records to prevent query failures when the history becomes extensive
- Improved Cowork progress accuracy: The execution process no longer collapses prematurely while waiting for subtasks, making task status clearer
- Improved feedback for Skill updates: Upgrade progress can now cover the entire application interface, with additional feedback for cancellation, security checks, and failures
- Improved text editing: Added a context menu to input fields with support for cut, copy, paste, and select all
- Fixed several bugs to improve system stability
Choose the version for your computer
- Added Team Edition features, including team management, usage analytics, and dedicated member quotas
- Added daily points rewards, allowing all users to claim points every day
- Integrated Skills and Connectors into a unified entry point for a more intuitive experience
- Added “Thinking Mode” and “Thinking Effort” options for the GLM-5.2 subscription model
- When exporting long conversations as images, the full conversation history is now loaded automatically, with progress and error notifications
- Upgraded browser annotations and file experience. Browser annotations are now retained as attachments and can be reopened from the Artifact panel
- Optimized the default interface, code font size, and side chat experience
- Increased the maximum number of custom model providers from 10 to 30
- Improved Windows installation and update recovery, with more accurate error messages
- Fixed several other bugs to improve overall application stability
Choose the version for your computer
- Improved the stability of Windows plugin installation and updates, preventing dependency links from being lost and protecting existing plugins if installation fails
- Fixed compatibility issues with application icons in the “Open With” menu on macOS and Windows, making the application list more reliable
- Fixed several other bugs to improve overall application stability
Choose the version for your computer
- Added “Thinking Mode” and “Thinking Effort” options for DeepSeek subscription models
- Introduced an all-new task progress view. Command execution, file read/write operations, reasoning, and media processing are now automatically categorized, with expandable details
- Made runtime status, elapsed time, and sub-agent progress clearer
- Improved file interactions, with sent attachments continuing to display as file cards
- Local file links now support choosing how to open files, saving as, copying content or file paths, and locating files in their folders
- Improved task management with dedicated indicators for scheduled task conversations
- Improved task execution stability
- Optimized the app startup and settings experience
- Improved stability when editing plugin configurations and closing dialogs
- Improved the reliability of Windows built-in runtime upgrades
- Fixed several bugs to improve overall application stability
Choose the version for your computer
- Added in-conversation search
- Improved mathematical formula rendering with support for more LaTeX formats
- Improved the stability of installation and updates on Windows
- Improved the stability of OpenClaw configuration syncing and plugin installation
- Fixed issues on Windows, including an unresponsive code block toolbar in full-screen mode
- Fixed several bugs to improve application stability
Choose the version for your computer
- Added an exclusive 5,000-point benefit for NetEase users
- Added an “Automatically Open File Previews” setting
- Optimized the AI engine startup process after login to reduce wait times
- Improved the stability of installation, updates, and uninstallation on Windows
- Fixed several bugs to improve overall application stability
Choose the version for your computer
- Fixed several bugs to improve application stability
Choose the version for your computer
- Added Side Chat, allowing follow-up questions at any time while the main task is running
- Added Kimi K3 support, compatible with built-in, custom, and subscription models
- Added sharing and deployment options for Artifact previews, making content publishing faster
- Optimized site quota handling, allowing new deployments after stopping older sites
- Optimized login recovery and automatic renewal, preventing accidental logouts during temporary network interruptions
- Added long-running model response notifications, providing more accurate timeout reason details
- Fixed some Windows installation and in-app update failures
- Enhanced email attachment download security to prevent writing files to unintended directories
- Optimized tool result handling in long conversations, improving task retry stability
Choose the version for your computer
- Built-in browser now supports multiple annotations and element style adjustments
- Improved file sharing and website deployment with clearer creation and permission management flows
- Windows now supports silent updates and automatic restarts, with enhanced recovery from installation failures
- Added a new creation entry point for AI Themes, with support for regenerating designs and automatically matching theme colors
- Added proactive protection for long-running tasks with oversized histories, reducing the risk of local AI engine crashes
- Fixed network interruptions being incorrectly marked as completed, with more accurate failure messages and retry handling
- Fixed occasional failures in configuration hot reloads, making model and runtime settings more reliable
- Fixed login retries, IM message flickering, conversation scrolling, and preview layout shifts
- Improved scheduled task dialogs and status displays, with proper line wrapping when editing model parameters
Choose the version for your computer
- Added a final reward claim entry and reward modal for the credits reset campaign
- Improved clipboard attachment parsing for better stability and testability
- Fixed several bugs to improve application stability
Choose the version for your computer
- Added desktop notifications for tasks awaiting confirmation or a response, with configurable timing for task completion notifications
- Added cleanup for temporary conversation files, with support for previewing and selecting directories to remove
- Improved the display of the reasoning process, presenting content in generation order while avoiding duplication
- Added support for GPT-5.6 and Grok 4.5 in custom model configurations
- Improved macOS update reliability by protecting the current version when installation fails and supporting recovery
- Enhanced Windows online installation and automatic repair capabilities, with improved compatibility with security software
- Fixed an issue where scheduled tasks could still be executed retroactively after being offline
- Fixed failures when copying Artifact content
- Fixed incorrect running status displays in IM conversations
- Fixed several bugs to improve application stability
Choose the version for your computer
- Added task guidance, allowing users to supplement or adjust requirements during execution
- Added multi-agent collaboration settings, with more complete subtask and execution history views
- Added support for attaching entire folders as context, making it easier to provide files in bulk
- Improved authorization confirmation and question-answering flows, with the option to temporarily collapse and resume them later
- Introduced a redesigned update card with clearer download, installation, and restart progress
- Improved Agent task pagination, sorting, and drag-and-drop reordering
- Improved local memory search to make memories from older Agents more accessible
- Fixed scheduled task delivery issues in DingTalk, WeCom, and Feishu group chats
- Fixed issues with multi-agent and IM conversation associations, Windows UI behavior, and input stability
Choose the version for your computer
- Added support for configuring multiple email accounts
- Added xAI as a model provider, allowing Grok models to be used with SuperGrok / X Premium quota
- Added MCP JSON configuration import
- Optimized scheduled tasks and notifications, with support for selecting a specific IM conversation or disabling notifications
- Optimized expert suites and skill management, adding an installed suites view and a “Use Now” entry point
- Improved memory management with support for multi-line entries, grouped display, and direct editing of original content, making long-term preference maintenance easier
- Optimized AI engine background heartbeat and long-conversation maintenance to reduce idle consumption, compression stalls, and list jumping
- Fixed an issue where the managed browser did not follow the system proxy/VPN settings
- Fixed several bugs to improve application stability
Choose the version for your computer
- Local service deployment sharing now supports Node service-based sites
- Added Goal Mode, with support for editing, pausing, resuming, and viewing status
- Optimized the Artifact panel to automatically open previews and display subtask results
- Added Qichacha MCP authorization to automatically configure enterprise-related services
- Added UI and code font size settings, with clearer interface prompts
- Added session diagnostic package export, with one-click repair for gateway startup failures
- Fixed issues with IM model sync, stuck context handling, and black screens on macOS
- Optimized the narrow-screen toolbar, subtask timestamps, and tool result display
- Improved the stability of OpenClaw caching, directory recognition, and tool startup
Choose the version for your computer
- Added an entry point for credit reset. Eligible users can quickly access and claim/use their benefits from the account menu or the homepage prompt.
- Optimized scheduled task loading and run history synchronization, reducing cases where the task list appears empty, history records cannot be opened, or outputs are lost.
- Fixed an issue where some OpenClaw built-in models could not be called properly when context limit configuration was missing.
- Optimized the custom model editing dialog, allowing users to scroll through and edit content when there is a large amount of information.
- Fixed an issue where the input toolbar could overlap when the Artifact panel was opened and resized.
- Fixed several bugs to improve application stability.
Choose the version for your computer
- Fixed an issue where Agent identity, personality, “About You,” and memory did not take effect in some tasks
- Optimized long-conversation cache stability to reduce repeated token consumption and improve continuous conversation stability
- Fixed issues with ChatGPT OAuth, MiniMax OAuth, and calls to certain Anthropic-compatible models
- Optimized error messages for credit limits, OAuth expiration, model permissions, network exceptions, and other issues, making the causes clearer
- Optimized scheduled task run history synchronization to reduce cases where task outputs and follow-up question records were lost
- Optimized the permission confirmation flow for plugins and commands to improve task execution stability
Choose the version for your computer
- Upgraded OpenClaw to v2026.6.1, improving Agent runtime stability and plugin compatibility
- Added Plan Mode, allowing users to review and confirm the execution plan before starting a task
- Added context caching support for Qwen and Claude models
- Improved Artifact, Mermaid, file sharing, and preview experiences
- Fixed several bugs to improve overall app stability
Choose the version for your computer
- Artifact sharing capabilities have been upgraded, supporting sharing for more file types, including Word, PPT, Excel, PDF, Markdown, and Mermaid.
- Improved the file preview experience for greater clarity.
- Upgraded the voice input experience.
- Improved long-conversation operations with a new “Scroll to Bottom” control, making it easier to quickly return to the latest content in long conversations.
- Fixed several bugs to improve app stability.
Choose the version for your computer
- Voice input has been upgraded to real-time recognition, converting your speech into text as you speak.
- Added the Computer Use expert toolkit on Windows, allowing LobsterAI to operate your computer after authorization.
- Sharing now supports public access.
- Improved fault tolerance for slow AI engine responses.
- Fixed the in-app update issue on Windows.
- Fixed several bugs to improve app stability.
Choose the version for your computer
- Data backup and migration: back up local data with one click and fully restore it on a new device or after reinstallation
- Task completion notifications: you’ll be proactively notified when a long-running task is finished
- Smoother login experience: after login, the window will automatically focus in the foreground
- Improved the visual presentation of Markdown formatting, code blocks, and the model selector in messages
- Fixed several bugs to improve app stability
Choose the version for your computer
- Added voice input
- Upgraded shortcuts with support for more quick actions
- Enhanced Artifacts preview: the preview panel can now expand into a larger view
- Improved document preview: PDF, Word, PowerPoint, and spreadsheet previews now support zooming; spreadsheet previews now support column width adjustment
- Added a new “Fix startup issues” option in Settings
- Fixed the Windows update installation flow to avoid compatibility issues caused by the legacy VBScript startup method
- Fixed several bugs to improve application stability
Choose the version for your computer
- Subscription users now support automated deployment and sharing of static pages
- MCP service management is more stable: npx-based MCP services now automatically optimize their startup paths, the service list now shows installation/optimization status, and retry is supported after failures
- @ media references in media generation are now resolved more reliably into actual file paths or URLs, reducing failures caused by incorrect reference images, first frames, or reference videos
- Batch deletion is now supported for subtasks/sub-agent sessions, and the corresponding gateway sessions will be cleaned up in the background after deletion
- Fixed several bugs to improve application stability
Choose the version for your computer
- Added Expert Kits, allowing users to install and use grouped capabilities with one click based on work scenarios
- Added Fork Conversation, allowing users to copy the current conversation history and start a new exploration branch
- Added Check for Updates to the plugin settings page, supporting manual detection and updates for npm and ClawHub plugins
- Added quick entries in the account menu for usage overview, top-up, inviting friends to earn credits, and more
- Added MiniMax M3 as a default model and updated the default context length for some custom models
- Large Markdown content and tool outputs now use a lightweight preview by default and can be expanded on demand, reducing page lag
- Optimized the video generation confirmation dialog to display consumption information more clearly, making it easier to confirm usage before submission
- Fixed an issue where the WeChat login state was lost after an overwrite installation on Windows. Future upgrades will no longer actively clear WeChat QR code credentials
- Optimized MCP runtime reuse and process cleanup, reducing background process buildup when using multiple conversations
- Fixed false reports of AI engine connection interruptions during large-output tool execution, and reduced unnecessary engine reloads during startup
Choose the version for your computer
- Added image generation capability, supporting Doubao-Seedream, Wan 2.7 image, and Minimax image series models
- Added video generation capability, supporting Doubao-Seedance 2.0, HappyHorse, Kling, and Minimax Hailuo series models
- Plugin market now supports syncing plugins from the OpenClaw extensions directory
- Added a tool loop detector to prevent aborted tool loops from continuing to consume tokens
- Fixed the issue where OpenClaw gateway sessions.patch timeout blocked chat.send
- Fixed the issue where GitHub Copilot token refresh triggered a gateway restart
- Fixed the issue where the sub-agent cleanup phase was blocked due to a hook failure
- Fixed several other bugs to improve application stability
Choose the version for your computer
- Custom parameters in JSON format can now be configured individually for each model, making it easy to control advanced settings like thinking depth
- The model's thinking process is presented in an independent module, clearly separated from the final response
- Sub-agent conversations are displayed independently in the sidebar, maintaining consistent styling and interaction with the main conversation
- The Artifacts preview browser now features more powerful annotation capabilities, facilitating direct markup and collaboration on artifacts
- The multi-instance bot management UI has been redesigned, making multi-account management clearer and more intuitive
- Added support for the Qwen 3.6 Plus Coding Plan
- Added browser network access mode selection and domain blocking configurations
- Fixed several bugs to improve application stability
Choose the version for your computer
- Added custom settings support for the model context window
- Added multi-tab support for the Artifacts right-side preview
- Added pairing code support for more IM platforms
- Upgraded the POPO plugin to 2.1.8
- Optimized the Dreaming
- Optimized the scheduled tasks
- Fixed compatibility issues with the Xiaomi Mimo model
- Fixed a bug caused by non-ASCII characters in the MCP service name
Choose the version for your computer
- Added context usage visualization, supporting manual compression of conversation context
- Added a plugin management panel, supporting online installation, uninstallation, and advanced configuration
- Significantly improved Excel and PPTX previews, supporting quick opening with local applications
- Reused the native OpenClaw MCP protocol for more stable and reliable external tool connections
- Reduced token consumption for heartbeat messages
- Images from tool execution results are now correctly displayed in the conversation
- Files generated in the conversation are uniformly displayed via the preview system for a more consistent experience
- Fixed an issue where the IM channel working directory was incorrect
- Fixed an issue where MCP servers failed authentication due to duplicate headers
- Fixed an issue where MCP stdio-type servers failed to start due to environment variable problems
- Several UI detail optimizations
Choose the version for your computer
- Added Dreaming memory consolidation feature, supporting background memory organization toggle, scheduling strategy, timezone, and model configuration
- Refactored Memory Settings page into a tabbed layout, added Dreaming status and diary content display
- Upgraded Youdao Cloud Notes Skill to v1.0.9
- Artifacts preview now supports Mermaid zoom controls, auto-refresh on external file changes, and hides the Code tab for binary files
- Added support for copying PNG/JPEG images to clipboard
- Improved code block component styling and fixed extra-long links not wrapping automatically
- Improved model selector placement and interaction styling
- Improved Skill selected state and main layout visual details
- Fixed IM channel configuration not taking effect after switching Agent
- Fixed Mermaid/PPTX preview anomalies and Ctrl+scroll zoom not working
- Fixed files with Chinese-character paths failing to open in browser on Windows
- Fixed some icons appearing blurry on low-DPI displays
- Fixed NO_REPLY tag at message tail not being correctly stripped during session sync
- Fixed bulk delete task operations in sidebar not taking effect
Choose the version for your computer
- Brand new Artifacts preview system, supporting real-time preview of HTML/SVG/React/Mermaid, as well as Office and PDF files
- Each Agent supports configuring an independent working directory
- Added Cron expression type for scheduled tasks, with paginated filtering supported for execution history
- Code highlighting engine upgraded to CodeMirror 6, supporting search and dark mode
- Assistant messages now display timestamp and model information
- Fixed an issue where characters were occasionally dropped when merging streamed text
- Fixed an issue where tool authorization prompts still appeared for stopped conversations
- Sidebar and overall UI optimizations
Choose the version for your computer
- Decoupled the main Agent's workspace from the working directory
- Added support for multiple POPO bots
- Upgraded the Youdao Note skill
- Fixed an issue where Markdown tables occasionally failed to render and table contents misaligned
- Fixed an issue where switching models in IM tasks did not take effect
- Fixed an issue where Qwen 3.6 Plus did not support image input
- Fixed an issue where WeCom dmPolicy / allowFrom configurations were hardcoded, preventing user configurations from taking effect
- Fixed occasional EPERM permission errors when deleting / installing skills on Windows
- Unified the success prompt for skill imports and stabilized the working directory during ClawHub imports
- Log optimizations
Choose the version for your computer
- OpenAI model provider supports ChatGPT OAuth login
- Xiaomi MiMo / Baidu Qianfan model providers support Coding Plan
- Upgraded OpenClaw-WeiXin to 2.1.10
- Automatically mask sensitive fields in logs
- Fixed unexpected Gateway restarts in scenarios such as plan model list updates, Qwen 3.6 Plus responses, WeChat QR code login, and homepage model switching
- Fixed issues when using DeepSeek V4 with custom providers
- Fixed issues including incorrect IM history sorting and garbled Chinese filenames in Feishu
- Increased startup timeout and fixed issues such as Windows installer exit code evaluation
- Fixed issues such as Cowork session deadlocks and file-in-use errors when deleting Skill files on Windows
Choose the version for your computer
- Added remote Embedding service configuration (supporting OpenAI, Gemini, etc.) for memory semantic search
- Fixed DeepSeek V4 thinking mode compatibility when tool calls are present (reasoning_content)
- Fixed issue where OpenAI native models become unavailable when system proxy is enabled
- Fixed issue where DiffView fails to render when edits array is passed to the edit tool
- Fixed Skill discovery path to align with actual runtime resolution path
- Fixed CJK (Chinese, Japanese, Korean) word segmentation issue in memory search on Windows
- Fixed old timeout callback incorrectly ending new conversation sessions
- Fixed Agent not correctly reporting the currently used model
Choose the version for your computer
- Added support for deepseek-v4-flash, deepseek-v4-pro, and kimi-k2.6 models
- Optimized skills and MCP search experience
Choose the version for your computer
- Discord and Telegram now support configuring multiple bot instances
- Added LM Studio provider to model configuration
- WeCom and POPO plugin upgrades
- Preserve draft input and attachments when switching sessions
- Fixed sharp native binding issue, added cfmind node_modules to NODE_PATH
- Fixed stale reply infinite loop when MCP tool times out
- Fixed incorrect request termination caused by MCP bridge listener object error
- Fixed some image-capable models unable to recognize image messages
- Windows Defender false positive fix
- Optimized chat.send RPC timeout issues
- Cowork content area UI width adaptive optimization
- Fixed enterprise configuration sync conflicts
- Fixed authentication UserProfile field inconsistency with actual API response
Choose the version for your computer
- Upgraded OpenClaw to v2026.4.14
- Added Lobster Email (clawEmail) IM channel
- Added a QR code scanning binding method for DingTalk bot configuration
- Automatically attempt to recover from backup during database exceptions, reducing the risk of data loss
- Added a welcome guide pop-up upon first launch
- Included OpenClaw gateway logs and daily log files when exporting the log ZIP package
- Optimized the client version upgrade experience
- Upgraded the NetEase POPO IM plugin to the latest version
- NetEase IM (NIM) now supports configuring multiple bot instances on the same account; added a QR code login guide process
- Optimized the main process startup sequence by displaying the window earlier and delaying Skill service initialization, significantly shortening cold start time
- Modified the OpenClaw Lark component to load only after HTTP listening is ready, reducing startup blocking
- Fixed the DingTalk dingtalk-connector startup crash by downgrading to the stable version 0.8.16
- Fixed an issue where the image path was still written to the prompt when a base64 image was already attached
- Fixed an issue where OpenClaw was rejected by security checks when its runtime path contained symbolic links
- OpenClaw heartbeat messages are no longer written to the user task history
- Fixed multiple IM message synchronization anomalies
- Expanded the Windows Defender exclusion scope to prevent the program from being mistakenly blocked
Choose the version for your computer
- Major OpenClaw Engine Upgrade (v2026.3.2 → v2026.4.8): The underlying AI execution engine has received a multi-version update, covering core improvements in tool-calling stability, multi-agent collaboration, task scheduling, and more.
- Gateway Startup Speed Significantly Improved: Disabled 64 unused built-in plugins, turned off the Bonjour service, and enabled esbuild minification, resulting in a substantially reduced startup time.
- Session Retention Duration Optimized: Chat sessions are now retained for 30 days by default, eliminating the need to re-establish context during continuous work.
- WeCom Supports Multiple Bots: WeCom (WeChat Work) can now integrate with multiple bots, meeting the diversion requirements of different business scenarios.
- POPO QR Code Login: Added a QR code scanning method to log in to POPO, streamlining the configuration process.
- Baidu Qianfan LLM Support: Added Qianfan integration, allowing direct usage within the model settings.
- MCP Tool Fixes: Resolved the issue where MCP tools were unavailable in the packaged version; changes to the MCP configuration now automatically trigger a restart to take effect.
- DingTalk Migrated to Official Connector: Connection is now more stable with better compatibility.
- Multi-IM Channel Adaptation: Comprehensive fixes and adaptations to the connection logic for platforms including Feishu, QQ, Telegram, Discord, and WeCom.
- WeChat Channel Restored: Fixed issues with WeChat QR code login and channel registration failures.
- Scheduled Task Fixes: Fixed issues such as cross-timezone task time drift, WeCom notification failures, and unsaved prompt logic.
- Conversation Stability: Fixed an issue where manually stopped conversations were unexpectedly resumed by delayed events; prevented duplicate error messages from being generated during gateway retries.
- Model Import: Fixed a crash that occurred when importing configuration files containing unknown Providers.
- Dark Mode UI: Fixed multiple dark mode display anomalies affecting elements like toggle switches and prompt cards.
Choose the version for your computer
- The MiniMax model provider authentication interface has been completely revamped, using Radio Cards instead of Tabs and displaying OAuth fields independently for a more intuitive configuration.
- Unified and integrated all model provider metadata into a shared registry, laying a solid foundation for adding new Providers in the future.
- Fixed an issue on Windows where incorrect path prompts during Skill creation caused files to be written to the wrong directory.
- Fixed an issue where the dirty state was not reset after saving a scheduled task, causing an erroneous "unsaved changes" prompt when leaving the page.
- Optimized the OpenClaw session keep-alive strategy, improving connection stability for long-running tasks.
- Corrected the default configuration values for Group Policy in the Feishu integration.
- Fixed an issue where outdated Provider configurations remained in openclaw.json after deleting all models.
- Optimized the Windows NSIS overwrite installation process, significantly accelerating installation speed by asynchronously deleting old directories.
Choose the version for your computer
- Support binding independent models and independent IM bots per Agent
- Support customizing message sending shortcuts in the settings panel and chat input box
- Added streaming output configuration support for the Feishu IM channel
- Added a "Skip missed scheduled tasks" toggle in the settings panel
- Support thumbnail previews and file type icons for attachments in conversations
- Add brand header and footer when exporting conversation images
- Marketplace interaction optimization - Search boxes, tabs, and category tags in Skills and MCP markets now support sticky scrolling (feat(skills,mcp))
- Improved the UX of the Agent settings panel (feat(agent))
- Added Host header validation and token authentication to CoworkProxy to prevent DNS rebinding attacks
- Fixed Bridge Server CORS configuration vulnerability and unvalidated Google Search URL scheme issue
- Fixed an issue where CRLF line endings on Windows caused patch files to fail to apply
- Fixed IM message history loss caused by reconcile during Gateway restart
- Fixed an issue where clipboard images were not written to disk, causing files to not be found in subsequent turns
- Correctly generate the complete OpenClaw configuration in enterprise mode
- Switched to using prevent-app-suspension instead of prevent-display-sleep to optimize standby behavior
- Fixed missing validation for required environment variables during MCP editing/installation
- Fixed a black screen issue caused by the useCallback hook order in the Agent model selection branch
- Detect and prevent duplicate shortcut bindings
- Added red asterisks to required fields in the scheduled task form
- Added an unsaved changes confirmation when canceling/returning from the scheduled task form
- Fixed an issue where the form state was not reset when reopening the AgentCreateModal
- Migrated from sql.js to better-sqlite3 and enabled WAL mode to improve performance
- Redux performance optimization: Used createSelector to split selectors, reducing unnecessary re-renders
Choose the version for your computer
- Support for multiple bot instances: DingTalk, Feishu, and QQ can now configure and run multiple bots simultaneously.
- More flexible scheduled tasks: Added an "hourly" schedule option, multiple workday selection, and the ability to directly specify the AI model and bot account within tasks.
- One-click conversation export: Chat histories can now be exported in Markdown or JSON format.
- More powerful skill library: Added a "create skills via conversation" feature and support for one-click importing of ClawHub skills.
- Significantly faster operations: The underlying database has been fully upgraded, resulting in a noticeable improvement in page response times and data query speeds.
- Smoother model integration: GitHub Copilot now supports automatic token refreshing; added guide links for obtaining API Keys across all models.
- Core bug fixes: Completely resolved the issue of lost attachments during multi-file uploads and added protection against duplicate conversation submissions.
- More stable system operation: Reduced unexplained engine restarts and optimized the reading stability of AI long-term memory.
Choose the version for your computer
- Fix the issue where, in certain circumstances, the non-enterprise configuration causes the sandbox mode to start, resulting in unavailability.
Choose the version for your computer
- Support for multiple custom model providers with automatic compatibility for existing configurations
- Added multiple new theme color schemes
- Added Diff view for editing tool calls
- Built-in Youdao Cloud Notes skill for direct read/write access to cloud notes in conversations
- Added agent icon selector
- Added built-in skill update functionality
- Added user community access
- Batch optimization of icon clarity on low-resolution screens
- Optimized preset agents to support multilingual display (Chinese, English, etc.)
- Fixed gateway restart loop issue under specific configurations
- Fixed corrupted log export archive issue
- Fixed unresponsive paste-and-send for standalone images
- Fixed Windows auto-start on boot failure
- Fixed memory failure issue
- Fixed blank page after deleting an agent
Choose the version for your computer
- Fixed several bugs to improve stability
Choose the version for your computer
- Fixed several bugs to improve stability
Choose the version for your computer
- Optimized content navigation bar
- Fixed several bugs to improve stability
Choose the version for your computer
- Multi-Agent Support
- Fixed several bugs to improve stability
Choose the version for your computer
- Fixed several bugs to improve stability
Choose the version for your computer
- Users will receive free model usage credits upon login.
- Launched the "Credit Top-up" feature, supporting recharges at any time to meet broader usage needs.
Choose the version for your computer
- Fixed several bugs to improve stability
Choose the version for your computer
- IM channels support Wechat
- Fixed several bugs to improve stability
Choose the version for your computer
- Support security detection for Skill installation and usage
- Enhance protection of OpenClaw-related sensitive information
- Performance optimization for Windows client
- Fix several bugs to improve system stability
Choose the version for your computer
- Fixed several bugs to improve stability
Choose the version for your computer
- MiniMax M2.7 model support
- Windows installer and startup optimization
- Scheduled task (Cron) UI optimization
- Fixed several bugs to improve stability
Choose the version for your computer
- Support loading OpenClaw native Skills
- IM channels support NetEase POPO
- Fixed several bugs to improve stability
Choose the version for your computer
- Switched the underlying Agent engine to openclaw
- Fixed several bugs to improve stability
Choose the version for your computer
- All-new Agent Engine
- This version is currently in Beta. It contains known bugs, and some features may experience instability.
Choose the version for your computer
- Fix several bugs and improve stability
Choose the version for your computer
- Add navigation support for tasks
- Optimize QQ IM's handling of file-type messages
- Fix several bugs to improve user experience
Choose the version for your computer
- IM channel supports WeCom
- IM channel supports QQ
Choose the version for your computer
- Optimize user experience
- Yunxin IM supports @mentions in group chats/circles
- Fix lagging and other bugs
Choose the version for your computer
- Support for the Skill Store
- Support for MCP
- Added Youdao Zhiyun as a model provider
- Fixed several bugs to improve stability
Choose the version for your computer
- Support formula rendering
- IM support for NetEase Bee
- Support image input for vision LLM
- Linux client packaging optimization
- Fixed bugs such as email connection issues on the Windows client
Choose the version for your computer
- Built-in Python environment on the Windows client
- Skills optimization & added version support
- Fixed bugs related to model providers
Choose the version for your computer
- Support automatic app updates
- IM scheduled task notifications support media sending and notification target persistence
- Added user manual
- Optimized the experience of using local models
- Fixed several bugs including kimi coding plan
Choose the version for your computer
- Support Kimi/GLM/Qwen/Volcengine coding plan
- Support media content input for DingTalk, Feishu, Telegram, and Discord
- Support custom enabling and disabling of system proxy settings
- Add Volcengine as a model provider
- Fix bug in skills parsing
- Fix OpenAI format incompatibility bug for MiniMax
- Fix bug where Windows installation paths did not support Chinese characters
- Fix issues in Windows environments regarding path format incompatibility with git-bash and internal command failures
Choose the version for your computer
- Added Seedance image generation skill
- Added Seedream video generation skill
- Added tech news skill
- Fixed IM, email, and other related bugs
Choose the version for your computer
- Added new IM channel: NetEase YunXin
- Added movie and music resource search skill
- Fixed imap-smtp-email skill
- Telegram bot supports User ID whitelist mechanism
Choose the version for your computer
- Model providers now support Xiaomi & custom model providers
- IM channel Telegram supports system proxy
- Fixed several bugs
Choose the version for your computer
- Fixed: Resolved timeout issues caused by modifying model API baseURL in sandbox
- Fixed: Fixed some bugs in model providers
Choose the version for your computer
- All-scenario Office Assistant — Data analysis, PPT creation, video generation, document writing, web search, email management, covering the entire daily office workflow
- Local + Sandbox Execution — Task execution supports direct local execution or isolated Alpine Linux sandbox
- Built-in Skills — Including Office document generation, web search, Playwright automation, Remotion video generation, etc.
- Scheduled Tasks — Support conversational initiation or GUI interface to add scheduled tasks, such as daily news collection, inbox organization, periodic report generation, etc.
- Persistent Memory — Automatically extract user preferences and personal information from conversations, remember your habits across sessions, the more you use, the better it understands you
- IM Remote Control — Trigger the Agent anytime via DingTalk, Feishu, Telegram, or Discord on your mobile device
- Permission Gate — All sensitive tool calls require explicit user approval before execution
- Cross-platform — macOS (Intel + Apple Silicon), Windows, combined with IM for mobile coverage
- Data Localization — SQLite local storage, chat records and configurations never leave your device