Skip to content

Releases: agno-agi/agno

v1.7.7

31 Jul 20:50
4022ea1

Choose a tag to compare

Changelog

New Features:

  • Revamped MCP integration: Our ****MCPTools and MultiMCPTools classes can now be initialized and used without an async context manager, providing a much easier experience. See the updated docs.
  • Morph Tools: Morph’s Fast Apply model as a Tool, it intelligently merges your original code with update snippets at 98% accuracy and 4500+ tokens/second.
  • Claude Interleaved Thinking: Added support for thinking in-between other content for Claude.

Improvements:

  • LiteLLM File & Image Understanding: Added support for file and image inputs for LiteLLM.
  • Upgrade ZepTools to v3 Zep: ZepTools are now compatible with Zep v3.

Bug Fixes:

  • OpenAIEmbedder() wrong dimensions for text-embedding-3-large: Automatically handles default dimensions length for text-embedding-3-small as 1536 and text-embedding-3-large as 3072.

What's Changed

New Contributors

Full Changelog: v1.7.6...v1.7.7

v1.7.6

24 Jul 14:35

Choose a tag to compare

Changelog

New Features:

  • Portkey Model Support: Added support for Portkey hosted models.
  • Bitbucket Tool: Added BitbucketTools with a variety of Bitbucket repository actions.
  • Jina Embedder: Added JinaEmbedder for using embedding models via Jina.
  • Row Chunking: Added RowChunking as a CSV chunking strategy.
  • EVM Toolkit: Added EvmTools to do transactions on EVM compatible blockchains using web3.
  • LinkUp Toolkit: Added LinkupTools for powerful search.
  • Background Execution Support for Workflows 2.0: Introduced background execution capabilities for Workflows 2.0, enabling non-blocking workflow execution with polling support. See docs here.

Improvements:

  • Async Bedrock Support: Added async execution support for the AWS bedrock implementation.
  • PostgreSQL Tools Updates: Various security and stability overhauls made to the PostgresTools toolkit.
  • Daytona Toolkit Updates: Added new tools for Daytona agent Toolkit

Bug Fixes:

  • LiteLLM Metrics: Fixed issue with metrics on streaming responses from LiteLLM.
  • Team Expected Output: Fixed issue where expected_output of members were overwritten by the team leader agent.
  • Workflows Async Generators: Fixed how async generator arun functions are treated. It now correctly keeps async generators as async generators and doesn’t convert it to a coroutine.
    • Before: Workflows with Async generator arun functions were incorrectly awaited as coroutines, which could cause runtime errors or prevent proper iteration through the yielded asynchronous values.
    • After: Async generator workflows are now properly recognised and handled as async generators, allowing for correct iteration over their yielded values using async for. This ensures all yielded results are processed as intended within asynchronous workflows.
  • LiteLLM Multiple Streaming Tool Calls: When Agno is run through LiteLLM against OpenAI chat models (eg. GPT4.1), multiple streamed tool_calls lost their individual argument streams. This has been resolved.

What's Changed

New Contributors

Full Changelog: v1.7.5...v1.7.6

v1.7.5

17 Jul 21:21
554a8d3

Choose a tag to compare

Changelog

New Features:

  • SurrealDB: Added SurrealDB support as a vector DB for knowledge bases.

Improvements:

  • Session Caching: Added cache_session attribute to allow users to switch off session caching, which improves on memory management.
  • Workflows 2.0 FastAPI Support: Added support for running the new workflows in FastAPIApp .

Bug Fixes:

  • Nested Tool Hooks: Fixed bug with nested tool hooks.

What's Changed

New Contributors

Full Changelog: v1.7.4...v1.7.5

v1.7.4

16 Jul 08:40
f0e1691

Choose a tag to compare

Changelog

New Features:

  • Workflows Revamped (beta): Our new Workflows implementation (internally referred to as Workflows 2.0) is a complete redesign of the workflow system in Agno, introducing a step-based architecture that provides better structure, flexibility, and control over multi-stage AI agent processes
    • Core Concepts:
      • Flexible Execution: Sequential, parallel, conditional, and loop-based execution
      • Smart Routing: Dynamic step selection based on content analysis or user intent
      • Mixed Components: Combine agents, teams, and functions seamlessly
      • State Management: Share data across steps with session state
    • For a comprehensive guide to the new Workflows system, check out the docs.
image
  • Pydantic Input for Agents/Teams: Both Agent and Team now accepts structured input (i.e. a pydantic model) on run() and print_response().

What's Changed

Full Changelog: v1.7.3...v1.7.4

v1.7.3

15 Jul 22:32
194d9b0

Choose a tag to compare

Changelog

New Features:

  • Session State on Run: You can now pass session_state when running an agent or team. See docs for more information.
  • GCS Support for PDF Knowledge Base: Added GCSPDFKnowledgeBase to support PDFs on Google Cloud Storage.

Bug Fixes:

  • Workflows Async + Storage: Fixed issues where sessions was not correctly stored with async workflow executions.
  • Session State Management: Fixed issues around session state management in teams and agents. Session state and session name will now correctly reset and load from storage if sessions are switched.
  • Metadata Support for Document Knowledge Base: Adds metadata support for DocumentKnowledgeBase
  • Session Metrics with History: Fixed bug with session metrics on Agent where history is enabled.

What's Changed

New Contributors

Full Changelog: v1.7.2...v1.7.3

v1.7.2

10 Jul 15:49
5547e4e

Choose a tag to compare

Changelog

New Features:

  • MySQL Storage: Added support for MySQLStorage as an agent/team/workflow session storage backend.
  • XAi Live Search: Added support for live search on the XAi model provider.
  • OpenAI Deep Research: Support added for o4-mini-deep-research and o3-deep-research models.

Improvements:

  • Memory Growth on Performance Evals: Added memory_growth_tracking as an attribute on PerformanceEval to enable additional debug logs for memory growth.
  • Agent/Team in Tool Hook: Added agent and team as optional parameters in tool hooks, for more flexibility.

Bug Fixes:

  • Gemini 2.5 Metrics: Fixed Gemini metrics to correctly include “thinking” tokens.
  • Claude tool calling: Fixed a bug related to parsing function call responses when using Claude models.
  • Team Metrics: Fixed a bug with team metrics when teams have history enabled.

What's Changed

New Contributors

Full Changelog: v1.7.1...v1.7.2

v1.7.1

04 Jul 08:11
e9d640f

Choose a tag to compare

Changelog

New Features:

  • Debug Level: Added debug_level to both Agent and Team. This is an int value of either 1 (default) or 2 (more verbose logging). Currently this only enables more verbose model logs, but will be used more widely in future.

Improvements:

  • Parser Model on Teams: Added parser_model support for Team. See docs here.
  • Support for Gemini Thinking: Added thinking_budget and include_thoughts parameters for Gemini model class.
  • Serper Tools: Made updates to the toolkit to include new tools search_news , search_scholar and scrape_webpage.
  • Valyu Tools: New Valyu toolkit for Deep Search capabilities of academic sources.
  • Oxylabs: Added OxylabsTools for adding more web-scraping capabilities to agents.

Bug Fixes:

  • DuckDB CSV parsing error: For CSV files use the custom read_csv method for improved CSV parsing
  • Full Team Metrics: Fixed an issue with calculation of the full_team_session_metrics on Teams.

What's Changed

New Contributors

Full Changelog: v1.7.0...v1.7.1

v1.7.0

26 Jun 18:26
ce752ca

Choose a tag to compare

Changelog

New Features:

  • Agent/Team Add Tool: Added convenience function to Agent and Teamadd_tool(tool) to append new tools after inititialisation.
  • Streaming Structured Output: Implemented structured output during streaming. This means streaming won’t be turned off when response_model is passed. The structured output itself is not streamed, but it is part of the iterator response when running an agent/team with streaming. The response model is set on a single RunResponseContentEvent and on the final RunResponseCompletedEvent.

Improvements:

  • Linear Teams Tool: Added tool to get the list of teams from Linear.

Bug Fixes:

  • Uppercase Structured Output: Resolved cases where pydantic model fields contain field names with upper case characters.

Breaking Change:

  • If you use run(..., stream=True) or arun(..., stream=True) on Agent or Team with a response_model set, the current behaviour would switch off streaming and respond with a single RunResponse object. After the changes mentioned above, you will get the Iterator[RunResponseEvent] / AsyncIterator[RunResponseEvent] response instead.

What's Changed

New Contributors

Full Changelog: v1.6.4...v1.7.0

v1.6.4

23 Jun 21:16
2059819

Choose a tag to compare

Changelog

New Features:

  • Brightdata Toolkit: Added multiple web-based tools via Brightdata.
  • OpenCV Video/Image Toolkit: Added tools for capturing image/video via your webcam.
  • DiscordClient App: Added a DiscordClient app for connecting your agent or team with Discord in the form of a discord bot.

Improvements:

  • FileTools File Search: Added search to FileTools.

Bug Fixes:

  • Fix User Control Flow with History: Fixed issues where user control flow (HITL) flows failed with message history.
  • Fix lance db upsert method for supporting knowledge filters: Fixed function upsert not using the parameter filters
  • Update mongo db hybrid search for filters: Mongodb now correctly uses filters for hybrid search
  • Fixed team.rename_session(...) that raises an AttributeError

Breaking Changes:

  • SerperApiTools: Refactored SerperApiTools to SerperTools for clearer naming.

What's Changed

New Contributors

Full Changelog: v1.6.3...v1.6.4

v1.6.3

18 Jun 10:05
6f79346

Choose a tag to compare

Changelog

New Features:

  • User Control Flows on Playground: The Agno Platform now support user control flows on the playground.
  • Team & Agent Events on RunResponse: Added store_events parameter to optionally add all events that happened during the course of an agent/team run on the RunResponse/TeamRunResponse.
  • Team Member Responses on Playground: The Agno Platform now shows member responses during team runs.
  • Behind-the-scenes on Playground: The Agno Platform now shows what is happening during runs, for both agents and teams.
  • Metadata filtering support for csv and csv_url knowledge bases: Add knowledge filters support for these knowledge base types.

Updates

  • Removed a prefix from async function names: asearch_knowledge_base, etc will now be the same as their sync counterparts when sent to the model. The names of functions are important for accurate function calling.

Bug Fixes:

  • AG-UI Fix: Fixed issue related to missing messages when using the Agno AG-UI app.
  • Chat History Fix: Fixed issue related to history not available when agent_id not set.
  • MongoDB ObjectId serialization issue when using with agent: Fixed issue while ****using mongodb vectordb with ObjectId present in the metadata it throws Object of type ObjectId is not JSON serializable

What's Changed

New Contributors

Full Changelog: v1.6.2...v1.6.3