Skip to content
Agents tracked: 501 Releases (7d): 777 Agent status: no reported issuesSince last limit reset: Claude 10d · Codex 8d For AI agents: Markdown Updated Nov 21, 2026

MetaGPT changelog: what's new each month

Every stable MetaGPT release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 8 months covered; the current month updates daily.

March 2025

1 release: 0.8.2

March brought bug fixes and merged improvements to MetaGPT.

March 2024

3 releases: 0.7.4 → 0.8.0

March brought improvements to the interpreter system, added support for new AI model providers, and enhanced testing infrastructure. The month included a major 0.8.0 release that unified naming conventions and expanded model compatibility.

Highlights

  • Added support for Gemini and proxy configuration
  • Introduced Qianfan and DashScope as new model providers
  • Unified Interpreter naming across the system
  • Enhanced testing with mocking for Mermaid ink and OpenAI embeddings

New

  • Gemini model provider support
  • Proxy support for model requests
  • Qianfan model provider integration
  • DashScope model provider integration
  • Unit test mocking for Mermaid ink rendering
  • Unit test mocking for OpenAI embeddings in DocumentStore and memory

Improved

  • Unified Interpreter naming conventions throughout the codebase
  • Updated Role.setactions() method in role.py
  • Upgraded ipykernel dependency to 6.27.1
  • Improved requirements.txt management

Fixed

  • Fixed GBK UnicodeEncodeError
  • Fixed action.llm.model configuration in debate example

February 2024

4 releases: 0.7.0 → 0.7.3

February brought significant architectural improvements to MetaGPT's configuration and context management system. The month focused on simplifying setup processes and restructuring core components for better flexibility and clarity.

Action needed
  • Configuration API changed from global CONFIG to Config instances.
  • Subscription concept replaced with tag-based subscription model.
  • Explicit serialize/deserialize functions removed from Role class.

Highlights

  • Simplified configuration system moving from global CONFIG to flexible Config instances for granular control.
  • Introduced Context objects that allow each Role to manage its own context independently.
  • New metagpt --init-config command provides a friendlier setup process for users installing via pip.
  • Subscription model changed to use tags for improved readability.
  • ActionNode can now be initialized directly from pydantic.BaseModel for easier structured definitions.

New

  • Context system for individual role-based context management.
  • metagpt --init-config command for simplified initial configuration.
  • Support for initializing ActionNode from pydantic.BaseModel.

Improved

  • Configuration system simplified to use Config instances instead of global state.
  • File repository restructured to consolidate FileRepo and GitRepo responsibilities.
  • WritePRD Action logic enhanced and buggy behaviour fixed.
  • Removed explicit serialize/deserialize functions from Role in favour of mixin approach.

Fixed

  • Numpy dependency error.
  • Human interaction bugs.
  • Code execution bugs.
  • Tool path bug.
  • Configuration error.

January 2024

5 releases: 0.6.0 → 0.6.6

January brought a major architectural overhaul of MetaGPT with version 0.6.0, moving to pure asynchronous operations and modernizing the codebase with Pydantic v2. The month focused on improving code quality, simplifying developer experience, and integrating with external platforms.

Action needed
  • All private variables converted to public as part of Pydantic v2 migration.
  • Codebase now requires Pydantic v2; code written for earlier versions must be updated.

Highlights

  • Moved all LLM API calls to pure asynchronous processing, removing synchronization code.
  • Upgraded entire codebase to Pydantic v2 for better type safety and performance.
  • Simplified example code dramatically—election debate now requires just 7 lines of code.
  • Test coverage increased from 30% to over 80%.
  • Added breakpoint recovery support for interrupted workflows.

New

  • Serialization and deserialization of important objects.
  • Breakpoint recovery to resume interrupted tasks.
  • Integration with Huggingface demo platform.
  • Integration with AgentStore for agent discovery and sharing.
  • Display of todo items and descriptions in AgentStore.

Improved

  • Simplified debate and knowledge base examples.
  • Provider lookup now uses default provider to prevent recursive searching issues.
  • Exception handling now maintains original exceptions from OpenAI and HTTPX libraries.

Fixed

  • Removed unexpected UserRequirement type message when no tasks remain.
  • Folder renaming now works correctly on Windows.
  • Recursive provider search issue resolved with DEFAULTPROVIDER.

December 2023

4 releases: 0.4.0 → 0.5.2

December brought major architectural changes to MetaGPT, including breaking updates to core APIs and experimental support for multiple languages and programming languages. The month focused on improving code generation stability and developer experience.

Action needed
  • Role.recv and Role.handle methods removed; use Role.observe instead.
  • Team.startproject renamed to Team.runproject.
  • startproject method now triggers deprecation warnings.

Highlights

  • Multilingual and multi-language programming support added in experimental form.
  • Incremental development capability introduced to support ongoing project evolution.
  • Core role communication APIs redesigned: Role.recv and Role.handle consolidated into Role.observe.
  • New aaskcode function enables more stable code generation through OpenAI function calling.
  • Support added for Zhipu ChatGLM API as an alternative model provider.

New

  • aaskcode function for stable code generation using OpenAI function calls.
  • Zhipu ChatGLM API support.
  • Incremental development mode (experimental).
  • CLI support (experimental).
  • Pip installation support.
  • Message subscription entry for roles.
  • GPT-4 Turbo and GPT-3.5 Turbo token counting.

Improved

  • Enhanced code review capabilities.
  • Role message handling unified into Role.observe method.
  • Default project root now uses current working directory in PyPI mode.
  • UserRequirement now watched by default when role is not explicitly configured.
  • Workspace initialization behavior improved.

Fixed

  • Workspace not existing issue.
  • Crashes due to OpenAI rate limit errors.
  • Problems in searchkb.py.
  • Dead loop when running startup example.
  • ContextVar OPTIONS LookupError.
  • Spelling errors in prompts.

November 2023

1 release: 0.3.0

November brought significant expansions to MetaGPT's capabilities, adding new specialized roles and actions while improving document storage and code execution features. The release focused on enhancing structured outputs and providing better tools for diverse workflows.

Highlights

  • LanceDB integration for improved document storage and retrieval
  • New open-interpreter tool enables code writing and execution within workflows
  • Structured JSON responses now enforced in PRD writing, API design, and project management
  • Multiple diagram rendering options reduce dependency on Node.js-based conversion
  • Three new specialized roles: TutorialAssistant, DetailMining, and PrepareInterview

New

  • TutorialAssistant role for educational content generation
  • DetailMining action for extracting detailed information
  • PrepareInterview action for software interview preparation
  • Open-interpreter integration for code writing and execution
  • OpenAI Moderation for content safety checking
  • Playwright, pyppeteer, and ink options for mermaid diagram conversion

Improved

  • Write PRD action now elicits JSON-formatted responses
  • Design API action produces structured JSON output
  • Project management actions return JSON-formatted results
  • Mermaid conversion defaults to Node.js mmdc but supports multiple backends
  • LanceDB replaces previous document storage for better performance

August 2023

1 release: 0.2.1

August brought expanded integrations, improved reliability, and a new QA engineer role to MetaGPT. The month focused on making the system work better in diverse environments and handling network issues gracefully.

Highlights

  • Added QA engineer role with test case writing, code execution, and debugging capabilities.
  • Integrated multiple web search engines and improved result parsing for better content retrieval.
  • Added timeout settings and network anomaly alerts for OpenAI connections.
  • Now runs on GitHub Codespaces and supports Qdrant vector database.
  • Fixed proxy support and improved default settings checking.

New

  • Integration with Qdrant vector database.
  • GitHub Codespaces support for running MetaGPT.
  • Multiple web search engine integrations including DuckDuckGo and Google API.
  • Enhanced web search result parsing for targeted content retrieval.
  • Timeout settings for OpenAI interface.
  • Network anomaly alerts for OpenAI connections.
  • QA engineer role with test case writing and code debugging abilities.

Improved

  • Web search result parsing now enables targeted content retrieval.
  • Fault tolerance added to cost and usage calculations.
  • Reduced code execution errors caused by network anomalies.
  • Default settings checking mechanism updated.
  • QA engineer role refined with additional capabilities.

Fixed

  • Fixed bug preventing proxy usage.
  • Improved handling of network anomalies in cost and usage calculations.

July 2023

1 release: 0.1.0

July brought major infrastructure improvements and expanded deployment options for MetaGPT. The release focused on making the platform more accessible and production-ready.

Highlights

  • Added support for Azure asynchronous API integration
  • Introduced separate output queue distinct from the message queue for better performance
  • Made gpt-3.5-turbo model available
  • Added Docker and pip installation support for easier deployment

New

  • Azure asynchronous API support
  • Docker deployment option
  • pip package installation
  • Separate output queue architecture

Improved

  • Separated output queue from message queue for improved message handling

Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. MetaGPT: pricing, features and alternatives · All changelogs