zz-haooo

zz-haooo / Meta-Team

Public

The implementation of the paper "Evolve as a Team: Collaborative Self-Evolution for LLM-based Multi-Agent Systems"

14
1
89% credibility
Found Jun 02, 2026 at 14 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Meta-Team is a research framework that lets teams of AI agents work together on complex tasks like software development, research, and automation. What makes it special is that after completing each task, the agents reflect on what went well and what didn't, then update their own behavior so they perform better on future tasks. Think of it like a work team that holds a quick retrospective meeting after each project and uses those lessons to improve - except the team members are AI agents. The system supports different types of teams optimized for different kinds of work, and includes built-in tests to measure how well the team improves over time.

How It Works

1
🔧 Set up your AI team

You connect your AI service account and choose a pre-built team of specialized agents for your type of work.

2
📋 Give your team a task

You describe what needs to be done - whether it's fixing bugs, writing code, researching topics, or automating workflows.

3
🤖 Watch your agents collaborate

The team lead assigns work to specialists, they communicate with each other, ask questions, and share results.

4
💡 The team reflects and improves

After finishing, agents discuss what worked well and what didn't - then update their own instructions for next time.

5
📊 See your results and team growth

You get the completed work plus a detailed report showing how your team performed and what they learned.

🚀 Your team gets smarter over time

Each task makes the whole team better - future work benefits from past experience.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 14 to 14 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is Meta-Team?

Meta-Team is a Python framework that lets teams of LLM agents not just execute tasks together, but actually improve themselves after each task completes. Based on a recent research paper, it implements three levels of self-evolution: agents update their own prompts and skills, communication patterns between agents get refined, and the team-level coordination rules get rewritten. The system runs on any OpenAI-compatible API and includes Docker-based sandboxing for safe code execution. You interact with it through a CLI that supports both single-task runs and an interactive mode where you can throw problems at your agent team one after another.

Why is it gaining traction?

The hook here is the "experience-driven evolution" concept. Unlike static agent pipelines that repeat the same failures, Meta-Team preserves execution context and runs a reflection phase after each task, turning what went wrong into reusable improvements. The framework ships with adapters for six major benchmarks (SWE-bench Pro, BeyondSWE, GAIA, LOCA-Bench, LoCoBench, and ResearchRubrics), so you can actually measure whether the evolution is working. The multi-scale approach is what separates it from simpler agent frameworks -- you're not just tweaking prompts, you're reshaping how the entire team coordinates.

Who should use this?

Researchers working on multi-agent LLM systems who want a ready-made evaluation pipeline and a framework for studying agent self-improvement. Developers building automated code repair or research pipelines who need agents that get better over time rather than plateauing. The steep setup (Docker dependencies, benchmark data downloads, API configuration) means it's not for casual experimentation -- this is for people running systematic experiments on agent teams.

Verdict

With only 14 stars and a credibility score of 0.9%, this is an early-stage research implementation, not production-ready tooling. The paper backing it adds legitimacy, but the codebase shows its experimental nature -- documentation is sparse and the benchmark adapters require significant manual data preparation. If you're evaluating multi-agent self-evolution academically, this is worth a look. If you need a reliable agent framework for production work today, wait for this to mature or choose something with a larger community.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.