Human-Agent-Society

Human-Agent-Society / CORAL

Public

multi-agent evolution organization

docs.coralxyz.com
19
0
100% credibility
Found Mar 19, 2026 at 19 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

CORAL runs teams of AI coding agents that collaborate to evolve better code solutions through experiments and shared knowledge.

How It Works

1
🔍 Discover a tough problem

You find CORAL and pick a coding challenge like optimizing a route or building a fast program.

2
📝 Share your starting point

Add simple starting code and a checker that scores how good solutions are.

3
🚀 Launch your AI team

Tell CORAL how many smart helpers to use and start them working together.

4
👥 Watch them team up

See agents experiment, share tips and notes, and beat each other's scores live on the dashboard.

5
📊 Track the leaderboard

Check improving scores, best attempts, and shared learnings anytime.

🏆 Grab the winning code

Download the top solution that's way better than where you started.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 19 to 19 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 CORAL?

CORAL builds organizations of autonomous AI agents—like Claude Code, Codex, or OpenCode—that collaborate on code evolution for optimization tasks. Feed it a seed repo and grader script, and agents spawn in isolated git branches, iterate solutions, eval via CLI, and share knowledge through notes, skills, and leaderboards. Python-based with commands like `coral start`, `coral ui` for dashboards, and `coral log` for tracking progress in claude multi agent github setups.

Why is it gaining traction?

It stands out in evolutionary multi agent systems by symlinking shared state across agents for real-time collaboration without sync delays, plus heartbeat prompts that trigger reflection or consolidation. Unlike solo agents or unrelated coral projects like coral edge tpu github, google coral github, or coral island github, CORAL handles multi-agent evolution with resume/stop, web monitoring, and builtin examples for TSP, kernels, and ML—making deep coral github experiments feel effortless.

Who should use this?

AI researchers tackling evolutionary multi-agent reinforcement learning in group social dilemmas, kernel/ML optimizers needing heuristic discovery, or teams prototyping self-improving codebases. Ideal for math conjectures like erdos or Kaggle comps where exhaustive search fails.

Verdict

Promising alpha for claude multi agent github workflows (19 stars, solid docs/examples), but 1.0% credibility score signals early days—test on toy tasks first. Grab it if you're into coral club-style agent societies; skip for production.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.