Changelog
V 3.5
Latest version release date: 2026/05/21
New Features
- Added English support
- Greatly improved the Office file operation experience
- Added support for conditional invocation
- Added support for connecting a WeCom bot
- Added support for custom output parameters
- Added support for configuring message notification policies
- Added private profiles
- Added slash-based quick Skill selection
- Tenant admins can now view each user's credit consumption
Changes
- The default model has been changed to DeepSeek-V4-Flash
- The
Generate Datatool is now disabled by default - Generating
.xlsxand.docxfiles now uses local tools
V 3.4
Release date: 2026/05/07
New Features
- Added support for purchasing a Pro subscription
- Added support for custom input parameters
- Added support for controlling available tools in free execution
- Added support for a custom workspace directory
- Enterprise users can now create API keys
- Added Skill switches
- Added manual question steps in step editing
- Added long-context compression
- Mac and Linux now support custom shells
- Built-in environment setup Skill for Python and Node, making it easier to support more operations. You can simply tell Xiaoji: help me use Skills to optimize the environment.
Changes
- The Print Page tool can now be called in conversations
- PPT generation is now provided as a Skill and the default PPT tool is disabled
- The shell tool is enabled by default
- In the enterprise edition, renewal gift credits are now monthly expiring credits
Experience Improvements
- Improved lag in very long conversations
- Accounts now come with email sending, image generation, and web search capabilities by default
- Added clickable local file links
V 3.3
Release date: 2026/04/07
New Features
- Added web remote control and conversation history viewing
- Added live progress view for WeChat remote control
- Added WeChat Pay credit purchase
- Added task pause/resume and the new
terminatedstate. See Task Management - Added the new tool: Visual Analysis
- Capabilities now support import and export
- Added user permission requests for dangerous operations
- Added WeCom group notifications for task results
- Added bit-Crew integration so MCP services configured in bit-Crew are loaded automatically in bit-Agent
- Added configuration of shell tool environment variables
Changes
- Cloud executor names are hidden during task execution
- Removed the Continue button and moved the Generate Capability button to the top
Record Pageis now disabled by default- Conversation-based schedule creation now only supports Free Execution schedules
Only available on specific pagesis currently an advanced parameter of theOpen Pagetool rather than a capability-level setting
Experience Improvements
- Improved small-screen display
- Upgraded visual design
- Improved shell command success rate and speed on Windows
V 3.2
Release date: 2026/03/17
Major update!
New Features
- Added support for remote control through WeChat
- Added support for reading and writing local files and shell operations. This feature operates on local files and runs commands, so it must be enabled manually.
- Added task execution options including AI takeover after failure and optional task notifications
- Added device management
- Added credits
- Added automatic memory updates
- Added double click and right click
- Added Print Current Page (currently supported only inside capabilities)
- Added Vision Exploration sub-agent (experimental) (Windows only for now)
- Added support for configuring Skills
- Tenant admins can add new tenant members
Experience Improvements
- Added Floating Window Mode
- The Upload File tool can now use a local path directly
- Removed the step limit
- Upgraded the tool selection bar
V 3.1
Major update!
Release date: 2026/02/26
New Features
- Added support for editing capability steps
- Added task result notifications
- Added Thinking Mode
- Added support for running tasks in the cloud
- Added the service marketplace
- Added configuration of available capabilities and services during exploration
Experience Improvements
- Added support for the latest Qwen-3.5 and GLM-5 models, and removed GPT-4o
- Removed the separation between exploration mode and general mode; all work can now be done in one mode
- Exploration steps are now embedded directly in the conversation
- Regular users can now configure MCP services in the console
- The step panel has been upgraded into a
session spacethat includes steps, files, and memory generated in the current conversation
Changes
- Web Page Notes now use wildcard matching, making them more flexible. Personal system prompts can now be changed from client settings.
- Clicking a link now behaves more like a real mouse click. Previously, links opened in the current page by default. As a result, old workflows that relied on
Go Backmay no longer work and should be changed toClose TaborSwitch Page.
V 2.7
Release date: 2025/11/20
New Features
Experience Improvements
- Conversation replay now supports playback speed adjustment and a draggable progress bar
V 2.6
Release date: 2025/11/04
New Features
- Added support for using capabilities in exploration mode
Experience Improvements
- Improved click logic for elements blocked by the floating window
- Improved capability step generation so AI repair is needed less often
- Improved MCP tool loading logic
- Added GLM and Qwen3-max as selectable models
V 2.5
Release date: 2025/10/14
New Features
- Added conversation sharing
- Added personal web page notes
- Added support for MCP services during exploration
- Added support for tenant-wide MCP services
- Added macOS support for the client
Experience Improvements
- Added dark mode to the console
- Improved some performance issues
V 2.4
Release date: 2025/09/02
New Features
- Added Schedule Management
Experience Improvements
- Improved download behavior and fixed garbled downloaded filenames
- Step screenshots now support download and copy
- Conversation history now supports categorization
- Added Linux client support
- Improved update logic
V 2.3
Release date: 2025/08/05
Console Features
- Added support for creating and dispatching tasks from the console
- Added support for editing capability information and added capability configuration items
- Added incognito execution for capabilities to reduce the impact of login state
- Added Web Page Notes configuration (tenant admin feature)
New Features
- Added the
Generate PPTactivity - Added the
Recognize CAPTCHAactivity - Added the
Generate Structured Dataactivity so capabilities can return structured results - Added the
Close Tabactivity - Added a visible tool list
Experience Improvements
- Generated HTML reports are more attractive
- Calling exploration from general mode now ends the exploration automatically, making it easier to chain exploration with other capabilities
V 2.2
Interaction Improvements
- Added highlighted operation positions in screenshots
- Added drag-and-drop file upload to the conversation bar
V 2.1
Release date: 2025/06/15
Interaction Improvements
- Improved the interaction page
- Added the ability to switch between step screenshots
- Exploration mode is now the default entry
- Improved the exploration-mode experience, with no need to open the browser in advance
- Improved installation experience
- Improved conversation experience
New Features
- Added conversation history
- Added automatic update support
- Added page generation, note taking, page navigation, and tab switching tools. See Tool List
- Improved activity selection for higher accuracy
- Greatly improved capability execution logic and speed
Compatibility
Important: after this version is released, V1 is no longer supported. Please upgrade as soon as possible.
V 1.1
Release date: 2025/04/03
New Features
- Added support for configuring MCP services
- Added general mode for more flexible browser control
- Added the error feedback entry so users can report issues
- Added settings with model selection and other options
- Added support for uploading files, currently including md, xlsx, docx, and txt
- Added voice input
- Added display of step image records
V 1.0
Release date: 2025/03/10
Core Features
- Added exploration mode
- Added conversion of exploration results into capabilities
- Added image upload support
Technical Features
- Agent-based architecture
- Natural language interaction
- Visual operation interface
System Requirements
- Supports mainstream operating systems
- Requires a network connection