MetaGPT Versions Save

🌟 The Multi-Agent Framework: First AI Software Company, Towards Natural Language Programming

v0.8.1

3 weeks ago

fix bug

v0.8.0

1 month ago

Release Highlights

Data Interpreter: Open Source and Better "Devin"

Data Interpreter is an autonomous agent that uses notebook, browser, shell, stable diffusion, and any custom tool to complete tasks.

Data Interpreter has achieved state-of-the-art scores in machine learning, mathematical reasoning, and open-ended tasks, and can analyze stocks, imitate websites, and train models.

It can debug code by itself, fix failures by itself, and solve a large number of real-life problems by itself.

We open-source our code and provide a wealth of working examples to give everyone access to state-of-the-art AI capabilities.

📝 Paper 🔗 Examples 📖 How to use

RAG

RAG module integrated! Common features such as indexing, retrieval, and rank are supported. Explore this example for a quick start.

New LLMs

New members for the supported LLM family: This time, Claude, QianFan, DashScope, and Yi. Check out here for configuration.

What's Changed

New Contributors

Kudos to all the contributors for making their first mark!

v0.7.7

1 month ago

A patch release for the following issues:

v0.7.6

2 months ago

A patch release for the following issues:

v0.7.4

2 months ago

A patch release for the following issues:

v0.7.3

2 months ago

A patch release for the following issues:

v0.7.2

2 months ago

A patch release for the following issues:

v0.7.1

2 months ago

A patch release for the following issues:

v0.7.0

3 months ago

Release Highlights

  • Easier and more flexible configuration: Simplified config setup, moving away from a global CONFIG to use Config instances for more granular control. Typical use cases involve assigning different LLMs to different Roles. Please check out here for detailed usage.
  • Introduced Context: Context is made up of config, repo, workspace, etc. Each Role can possess its own context, allowing more flexible context management.
  • Added metagpt --init-config command: A more friendly setup process for pypi package users.
  • Repo restructuring: Consolidated FileRepo and GitRepo responsibilities to simplify code and logic.
  • Subscription to tags: Changed the concept of subscription to tags for improved readability and usability.
  • Removal of explicit serialize/deserialize functions: Removed these functions from Role, used mixin instead for better clarity.
  • Supported initializing ActionNode from pydantic.BaseModel: An easy and structured way to define ActionNode.
  • WritePRD Action restructuring: Simplified code, enhanced effectiveness, and fixed buggy logic.
  • Env restructuring (experimental): Re-abstracted Env to accommodate various environments more effectively; Uses mark_as_readable mark_as_writable to decorate APIs to uniformly observe and act on the environment through observe and step.
  • Embraced Multi-modality: Added a vision agent based on GPT-4V, check out the example here!
  • Improved Incremental development: Higher success rate of incremental development compared to previous version.
  • Introduced CodeInterpreter: A general agent who solves problems through coding. The agent leverages on plans and tools, capable of tackling a wide range of problems, including data visualization, machine learning modeling, math reasoning, web scraping, webpage imitating, etc. We are adding examples here!

What's Changed

Backwards Incompatible Changes

  • IMPORTANT: We removed config.yam, key.yaml and switched to a minimal config2.yaml. Please check the latest setup method here
  • We removed Role._init_actions and added Role.set_actions, which allows actions setup at any time besides initialization. For Role initialization specifically, users should also use self.set_actions(...) now

Features

  • Easier and more flexible configuration, introduced Context, embraced Multi-modality, improved Incremental development, introduced CodeInterpreter, and more (see release highlights) #852 by @geekan @mannaandpoem @garylin2099 @shenchucheng @voidking @iorisa @better629 @stellaHSR @orange-crow @buttogo @FromCSUZhou @luxiangtaoya
  • Supported Generating class views from source code #676 by @iorisa

Bugfixes and improvements

  • Improved testing SOP, increased test coverage, and fix bugs during testing, ensuring quality and preparing for more effective development. #673 #678 #685 #688 #689 #690 #694 #696 #698 #702 #703 #707 #709 #710 #711 #713 #715 #718 #727 #729 #863 #867 by @shenchucheng @voidking @iorisa @garylin2099
  • Fixed is_human not working at Role initialization #715 by @kkdev163
  • Fixed system_prompt error at BaseLLM #732 by @better629
  • Constrain the language for the qa_engineer #743 by @arnaudgelas
  • Fixed: requirements.txt not written to the disk #744 by @arnaudgelas
  • Fixed multiple issues at qa_engineer.py, including crashing if no code_doc, generating tests for non-python files, trying to install dependencies when there are none #765 #775 by @arnaudgelas
  • Upgraded Zhipu API #770 #861 by @better629
  • Fixed prompt logic when defining to whom the message should be sent. #785 by @arnaudgelas
  • Fixed JSONDecodeError when parsing LLM outputs #799 by @HuZixia
  • Modified get_choice_delta_text for safer dict access while using ZHIPU API #801 by @pkdog

v0.6.6

3 months ago

A patch release for the following issues: