Release history
beeai-framework releases
Build production-ready AI agents in both Python and Typescript.
All releases
70 shown
Minor fixes and improvements.
Full changelog
0.1.29 (2026-05-05)
- bump deps
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/typescript_v0.1.28...typescript_v0.1.29
- Restricted PEP 621 metadata Python version to <3.14
- serve: Zed ACP adapter for exposing agents over stdio
- mcp: exclude_none parameter now configurable in MCPTool
Full changelog
Bug Fixes
- MCP output types, requirement target extraction (#1435)
- restrict Python version to <3.14 in PEP 621 metadata (#1433)
- deep copy PromptTemplate config to prevent shared mutable state across agents (#1419)
Features
- serve: add Zed ACP adapter for exposing agents over stdio (#1428)
- mcp: make exclude_none configurable in MCPTool parameters (#1429)
- validate class names in dynamic adapters (#1408)
New Contributors
- @SoftEgLi made their first contribution in https://github.com/i-am-bee/beeai-framework/pull/1419
- @iron-prog made their first contribution in https://github.com/i-am-bee/beeai-framework/pull/1409
- @RinZ27 made their first contribution in https://github.com/i-am-bee/beeai-framework/pull/1408
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.79...python_v0.1.80
## Features - update dependencies to fix security issues
### Features * update dependencies to fix security issues
## Features - update `langchain` and related dependencies to fix CVE issues
- VercelAI SDK v6 integration
- Improved Groq error handling
- Extended parameters for VectorStore and VectorStoreSearchTool
Full changelog
Features
- extend VectorStore and VectorStoreSearchTool parameters
- update agentstack-sdk to the latest version (#1405)
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.76...python_v0.1.77
- Agent Stack RAG integration with AgentStackVectorStore and AgentstackEmbeddingModel adapters
- Auto-recovery from malformed tool calls and empty responses
Full changelog
Features
- backend: auto recover from malformed tool calls and empty responses (#1388)
- adapters: add Agent Stack RAG capabilities (AgentStackVectorStore, AgentstackEmbeddingModel) (#1384)
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.75...python_v0.1.76
### Features * auto recover from malformed tool calls and empty responses
- Middleware support in ChatModel
- isEmpty helper for StreamToolCall
Fixes event propagation and tool call handling in agents and adapter settings.
- LiteAgent for efficient agent implementations
- allow_prompt_caching flag for ChatModel class
- Agent performance improvements
Full changelog
Bug Fixes
- adapters: remote AgentStack server connection (#1360)
Features
- agents: add LiteAgent (#1355)
- backend: add allow_prompt_caching flag to the ChatModel class (#1368)
- agents: performance improvements
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.73...python_v0.1.74
- RequirementAgent
- GlobalTrajectoryMiddleware
- Execution interception support
- ToolCalling agent is deprecated
- Bedrock adapter: support for top-level union operators removed
- Improved change propagation in AgentStackChatModel
Full changelog
Bug Fixes
- adapters: disable support for top-level union operators in Bedrock
Features
- agents: deprecate the ToolCalling agent (#1354)
- adapters: improve change propagation in
AgentStackChatModel(#1350)
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.72...python_v0.1.73
- Cost and token usage information in agent outputs
- API key support for Bedrock adapters
Full changelog
Bug Fixes
- agents: handle prompt caching edge cases in the RequirementAgent with Bedrock
- adapters: update the AgentStack form extension (#1344)
Features
- agents: add cost and token usage information to the output
- adapters: allow passing an API key for Bedrock
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.71...python_v0.1.72
- Prompt caching strategy applied to RequirementAgent
- Default parameters added to AgentStackAgent factory method
- Option to disable A2A streaming in adapters
Full changelog
Bug Fixes
- logger: use BEEAI_LOG_LEVEL env variable (#1332)
Features
- agents: apply prompt caching strategy to RequirementAgent (#1340)
- adapters: update AgentStack sdk
- adapters: add default parameters for AgentStackAgent factory method (#1336)
- adapters: add option to disable A2A streaming (#1330)
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/typescript_v0.1.21...python_v0.1.71
- BeeAIPlatform adapter renamed to AgentStack
- Health-check support added to serve endpoint
Full changelog
Features
- upgrade all dependencies (#1322) (cb42104), closes #1320
- serve: add health-check support (#1235) (123232a), closes #1194
- adapters: rename BeeAIPlatform to AgentStack (#1256) (366e473)
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/typescript_v0.1.20...typescript_v0.1.21
- Improved error handling for Groq adapter
Full changelog
Bug Fixes
- adapters: langchain text_splitter types and refactor
Features
- adapters: improve error handling for Groq
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.69...python_v0.1.70
- Human-in-the-loop support via `io_confirm` adapter
- Qwen and DeepSeek model provider support
- A2A adapter health checks
Full changelog
Bug Fixes
- adapters: propagate changes to the AgentStack Chat model (#1308)
Features
- agents: refactor and improve the RequirementAgent (#1317)
- adapters: add
io_confirmfor human-in-the-loop support (#1306) - adapters: make default exports lazy for AgentStack (#1316)
- adapters: make default exports lazy for A2A
- adapters: add health check support for A2A (#1235)
- adapters: propagate the AgentStack client (#1315, #1325)
- adapters: add Qwen and DeepSeek providers (#1281)
- deps: bump
ddgs - deps: switch from
llama-indextollama-index-core
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.68...python_v0.1.69
- Support for handling streamed tool calls without an ID
Full changelog
Bug Fixes
- backend: fixed cloning of ChatModel/EmbeddingModel (#1311)
Features
- backend: added support for handling streamed tool calls without an ID (#1312)
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.67...python_v0.1.68
- Removed `next_step` parameter from ThinkTool
- Backend: auto fix for malformed tool calls
- Agents: rules now support a `reason` field when using requirements
- Middlewares: custom message formatter for GlobalTrajectoryMiddleware
Full changelog
Features
- backend: auto fix malformed tool calls (#1291)
- agents: add a
reasonto a rule when using requirements (#1289) - tools: remove
next_stepfrom the ThinkTool (#1303) - middlewares: add a custom message formatter to the GlobalTrajectoryMiddleware (#1287)
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.66...python_v0.1.67
- Add method to create OpenAPITool from URL
- Improve cloning mechanism
- Handle incomplete Unicode chunks while streaming
Full changelog
Bug Fixes
- backend: extract content from the last corrupted chat output chunk
Features
- tools: add method to create OpenAPITool from URL (#1284)
- tools: improve cloning mechanism (#1283)
- backend: handle incomplete Unicode chunks while streaming (#1279)
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.65...python_v0.1.66
- Nested trajectory support for AgentStack
- Improved typings for ConditionalRequirement
- Allow passing custom URL for A2A Client
Full changelog
Bug Fixes
- adapters: allow passing custom URL for A2A Client (#1276)
Features
- adapters: add nested trajectory support for AgentStack (#1273)
- agents: improve typings for ConditionalRequirement
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.64...python_v0.1.65
- Host parameter added for A2A adapters (#1269)
- Memory propagation to context in ReactAgent
- GlobalTrajectoryMiddleware now emits events (#1272)
Full changelog
Bug Fixes
- tools: correctly clone the HandoffTool
Features
- adapters: add host parameter for A2A (#1269)
- tools: sanitize name attribute in the HandoffTool
- agents: propagate memory to the context in ReactAgent
- tools: improve error handling in HandoffTool
- middlewares: emit events from GlobalTrajectoryMiddleware (#1272)
- middlewares: propagate updates from StreamToolCallMiddleware without streaming
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.62...python_v0.1.64
- Filtering agents by state in AgentStack
- PromptTemplate exposure in MCP server
- aserve method added to MCPServer
Full changelog
Bug Fixes
- adapters: prevent importing extra dependencies when using the OpenAI Chat model (#1263)
Features
- dependencies: relax the version constraint for the cachetools dependency
- adapters: allow filtering agents by state in AgentStack (#1258)
- adapters: support exposing PromptTemplate in the MCP server (#1260)
- adapters: add the
aservemethod to MCPServer
Full changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.61...python_v0.1.62
- Update code to use AgentStack instead of BeeAIPlatform in adapters
- BeeAIPlatform renamed to AgentStack in adapters module
- Settings extensions for AgentStack agents
Full changelog
Features
- adapters: rename BeeAIPlatform to AgentStack (#1256)
- adapters: add Settings extensions for the AgentStack agents (#1223)
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.60...python_v0.1.61
- Streamable-http now enabled for MCPServer
Full changelog
- Streamed tool call parsing with stream_stable=True
- Updated BeeAI SDK
Full changelog
Bug Fixes
- adapters: handle missing messages in BeeAIPlatform (#1237)
- adapters: add missing meta property for A2A/BeeAI Platform Agents (#1243)
- adapters: correct decorator order in ChatToolFunctionDefinition (#1234)
Features
- adapters: update BeeAI SDK (#1231)
- middlewares: parse streamed tool calls with stream_stable=True
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.58...python_v0.1.59
### Bug Fixes * **adapters:** fix MCP server logging
- OpenAI Chat Completion/Responses serve module
- Provider-specific parameters propagation in BeeAIPlatformChatModel
- New parameters added to A2A Agent
Full changelog
Bug Fixes
- serve: properly serialize tool output when serving via MCP (#1219)
- adapters: properly handle provider-specific settings in BeeAIPlatformChatModel
- adapters: handle unsupported JSON Schema keywords (Groq) (#1216)
Features
- adapters: add OpenAI Chat Completion/Responses serve module (#1182)
- adapters: propagate provider-specific parameters in BeeAIPlatformChatModel (#1225)
- adapters: add new parameters to the A2A Agent (#1224)
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.57...python_v0.1.58
- A2A server context support in adapters
Full changelog
Bug Fixes
- agents: tool visibility regression in RequirementAgent (#1204)
- emitter: handle cleanups when cloning (#1211)
- emitter: class method reference matching (#1209)
Features
- adapters: add A2A server context (#1207)
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.56...python_v0.1.57
- Tool choice mapping added to all providers in the platform
- LangChainChatModel clone method updated
Full changelog
Bug Fixes
- backend: remove property decorator from computed_field (#1191)
- agents: prevent duplications of tokens while streaming an answer (#1188)
Features
- adapters: add tool choice mapping to all providers in the platform (#1185)
- adapters: update clone method in LangChainChatModel (#1189)
- RequirementAgent streams final answers
- Backend supports additional message content types
- RequirementAgent propagates retry parameters to LLM
Full changelog
Bug Fixes
- adapters: sets a default timeout for the A2A client (#1184)
Features
- agents: streams the final answer in the RequirementAgent (#1178)
- agents: propagates retry parameters to the LLM in the RequirementAgent
- backend: supports additional message content types (#1150)
- backend: updates Uvicorn version (#1180)
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.53...python_v0.1.55
- Dynamic registration of BeeAI Platform extensions
- Tool metadata support for BeeAIPlatformServer
- Listener priority configuration in emitter
Full changelog
Bug Fixes
- internals: handle processing of more complex schemas (#1168)
Features
- adapters: support dynamic registration of BeeAI Platform extensions (#1162)
- adapters: add tool metadata to BeeAIPlatformServer (#1166)
- emitter: allow setting listener priorities (#1170)
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.52...python_v0.1.53
- HandoffTool now accepts any runnable
Full changelog
Bug Fixes
- internals: serialization of lists/sets (#1161)
- tools: adjust typings for MCPTool to accept
streamablehttp_client - backend: prevent propagating internal parameters to the request
Features
- tools: allow HandoffTool to accept any runnable (#1158)
Full Changelog: https://github.com/i-am-bee/beeai-framework/compare/python_v0.1.51...python_v0.1.52
- RequirementAgent moved from experimental module — verify and update import paths
- Structured output validation via validate_response_format flag in ChatModel
- Transformers ChatModel adapter for expanded model family support
- RequirementAgent graduates from experimental module
Full changelog
Bug Fixes
- adapters: correctly propagate Google VertexAI parameters (#1146)
Features
- agents: move RequirementAgent out of the experimental module (#1143)
- adapters: add support for Transformers ChatModel (#1087)
- backend: add the
validate_response_formatflag for structured outputs in ChatModel (#1156) - adapters: add serve factories for Runnables (#1139)
Full Changelog: Compare v0.1.50...v0.1.51
MCP server outputs flattened for cleaner, more consistent results.
Adds A2AAgent and A2AServer adapters with beeai-sdk integration.
ChatModel refactored to Runnable interface; create() renamed run(), create_structure() merged.
Adapters now support nested initialization of BeeAIPlatformChatModel.
Adds adapter support for BeeAIPlatform server/client extensions and LangChain ChatModel integration.
- Agents require positional input parameter (str | list[AnyMessage])
- Agent return types renamed (AgentRunOutput → AgentOutput, etc.)
- run() method signature changed
- Runnable interface unification
- Parallel agent runs support
- Streamable-http MCP support
- Interception support
### Bug Fixes - match nested events in GlobalTrajectoryMiddleware (#1057)
Adds retry logic to ChatModel for improved reliability.
BeeAIPlatform agent and server switched to beeai-sdk and A2A.
Introduces runnable interface to unify execution across framework components.
- Parallel tool calls
- Dynamic document processor loading
- Backend text splitting class
Bug fixes for A2A and MCP adapters, plus tool duplication in Requirement Agent.
Adds XAI (Grok) adapter with chat and embedding support.
Introduces RAG module and RAG agent for retrieval-augmented generation workflows.
Introduces Groq adapter embedding for expanded model provider support.
- IBM watsonx Orchestrate integration
- oneOf/anyOf schema support
- Context-dependent IO support
- force_prevent_stop attribute
- Improved callback invocation
RemoteAgent refactored to ACPAgent and BeeAIPlatformAgent with MCP serve integration.
Experimental RequirementAgent combines LLMs, tools, and requirements in declarative interface.
Add MCP and BeeAIPlatform serve integrations, replacing RemoteAgent.