v0.14.0

What's Changed

• Add controller args support by @mlikasam-askui in #127
• chore: remove unused askui hub/workspaces api client by @adi-wan-askui in #128

🚀 Features

• Controller Arguments Support: Enhanced controller functionality with configurable arguments support, providing better flexibility and control over AskUI Controller, e.g., showing/hiding overlay

🚨 Breaking Changes

• AskUI Hub/Workspaces API Removal: Completely removed the unused AskUI Hub and Workspaces API client (~25k lines of code) that led to problems with long paths on Windows
  • Removed AskUIHub class and all workspace-related APIs
  • Removed hub property from AgentToolbox class in src/askui/tools/toolbox.py:33
  • Removed all AskUI workspaces exceptions from src/askui/tools/askui/exceptions.py

Full Changelog: v0.13.0...v0.14.0

v0.13.0

What's Changed

• feat(excel): add support for Excel file processing in VisionAgent by @danyalxahid-askui in #121
• feat: mcp tool support by @adi-wan-askui in #122
• Fix/persistence in chat api by @adi-wan-askui in #123
• feat/create thread and run endpoint by @adi-wan-askui in #124

🚀 Features

• Office Document Support: Excel and Word: Pass .xlsx, .xls, .docx, .doc files to <Agent>.get() as string (path) or pathlib.Path
• MCP (Model Context Protocol) Tool Support:
  • Pass fastmcp.Client to ToolCollection and ToolCollection to <Agent>.act() to use MCP tools
  • Use MCP config endpoints of Chat API to configure MCP servers to be used by agents within chat (for now, no filtering of servers or tools, unavailable servers are ignored)
• Chat:
  • Start agent run / conversation with single create_thread_and_run endpoint to create a thread (conversation) and execute an agent to continue the thread (conversation)
  • Add Files API for managing files referenced from within chat messages
  • Scope resources to workspaces instead of sharing them across all workspaces (with exception of assistants/agents)

🐛 Bug Fixes

• Chat
  • Consistent encoding of files used for storage of resources (UTF-8)
  • Resource ID generation and ordering of resources when listing them

Other

• Unified input source types: Replaced various type definitions (Img, Excel, Pdf) with standardized InputSource type
• Improved documentation: Extracted documentation from README to dedicated /docs folder
• Chat message storage optimization: Changed from single JSONL file to individual files for better scalability
• Async/await: Switch to async/await under the hood and for creating a run for supporting async-first fastmcp framework and improving performance
• New documentation: Added detailed guides for MCP, extracting data, chat API, and more

🚨 Breaking Changes

• ActModel.act(): The tools parameter is now a ToolCollection instead of a list of Tool
• Chat: You are going to loose all your chats when upgrading to this version as we changed how we store them

Full Changelog: v0.12.1...v0.13.0

v0.12.1

What's Changed

• Update grpcio to >=1.73.1 by @onur-askui in #120
• Feat: Add locate all function by @mlikasam-askui in #118

🚀 Features

• feat: AgentBase.locate_all() to locate (points) of all element matching a locator by @mlikasam-askui in #118

🐛 Bug Fixes

• fix: solve gRPC version conflict by @onur-askui in #120

Full Changelog: v0.12.0...v0.12.1

v0.12.0

What's Changed

• fix encoding issue by @mlikasam-askui in #117
• fix: change ModelName to a regular class so that (class) constants can be used like regular strings in Python >3.10 by @adi-wan-askui in #116
• feat(get): support PDF processing and refactor image handling by @danyalxahid-askui in #115

🚀 Features

• feat(get): support PDF processing and refactor image handling by @danyalxahid-askui in #115
  • Updated the get method to accept both images and PDFs as sources (size up to 20MB)
  • Renamed the image parameter to source for clarity
  • Added support for PDF processing in ASKUI-hosted gemini models (default)

🐛 Bug Fixes

• fix: change ModelName to a regular class so that constants can be used like regular strings in Python >3.10 by @adi-wan-askui in #116
  • Fixes issues with using Gemini where "ModelName.GEMINI__2_5__FLASH" was passed as model id instead of "gemini-2.5-flash"
• fix: encoding issue by @mlikasam-askui in #117

🚨 Breaking Changes

• Renamed parameter of AgentBase.get() method from image to source
• Data URL (source of AgentBase.get()) now requires the mimetype to be specified (e.g., "data:application/pdf;base64,JVBERi...")

Full Changelog: v0.11.0...v0.12.0

v0.11.0

What's Changed

• Re release of the previous release 0.10.5 as a major upgrade because of breaking changes

Full Changelog: v0.10.5...v0.11.0

v0.10.5

What's Changed

• Tars improvements by @stephenkopylov in #101

🐞 Bug Fixes

• Issue with Tars Model Initialization

🚨 Breaking Changes

• UI-TARS: If you want to use UI-TARS, it is from now on required to provide a model name using the env variable TARS_MODEL_NAME.

Full Changelog: v0.10.4...v0.10.5

v0.10.4

What's Changed

🐞 Bug Fixes

• fix(retry): update on_exception_types type hint and add support for m… by @danyalxahid-askui in #113

Full Changelog: v0.10.3...v0.10.4

v0.10.3

What's Changed

• refactor(gemini): retry on more error codes & interpret retry-after header by @adi-wan-askui in #112

Full Changelog: v0.10.2...v0.10.3

v0.10.2

🐞 Bug Fixes

• Optional dependencies (regression): Fixed that importing VisionAgent failed when installing askui via pip install askui as optional dependencies were imported although not available (made fastapi a required dependency as some tools require it)

Full Changelog: v0.10.1...v0.10.2

v0.10.1

What's Changed

• test: add test for get with deeply nested response schema with model that does not support recursion by @adi-wan-askui in #110
• fix: rm static dependencies on optional dependencies by @adi-wan-askui in #109
• fix/retry askui api by @adi-wan-askui in #111

🐞 Bug Fixes

• Optional dependencies: Fixed that importing VisionAgent failed when installing askui via pip install askui as optional dependencies were imported although not available
• Stability: Made AskUI model usage more stable, 502 errors from API are now also retried and retry logic is improved with it so we retry 3 times (30, 60, 120 s)

Full Changelog: v0.10.0...v0.10.1