Human Agent Society
Human Agent Society is an open-source AI-agent research and engineering organization/project behind CORAL, an autoresearch framework for autonomous coding agents. CORAL coordinates long-running agents in isolated Git worktrees, continuously grades experiments, and shares discoveries through persistent memory so teams can search for better solutions over time.
At a Glance
- AI research labs and academic researchers
- ML engineers and data-science teams
- Systems and GPU-kernel engineers
- Developers building autonomous coding-agent workflows
- +1 more
AI Tools by Human Agent Society
(1)
CORAL
Open Source Multi Agent Framework
Discussions
No discussions yet
Be the first to start a discussion about Human Agent Society
Latest News
Reef v0.1.2 released with inference-record, artifact-review, and harness-session fixes
Reef v0.1.0 released with evolution budgets, credential scanning, release identity, and review-gated releases
CORAL v0.7.21 released
Reef v0.0.2 released and tagged builds made available
Products & Services
Open-source autoresearch infrastructure powered by autonomous coding agents. It creates isolated Git worktrees, runs continuous grading, maintains shared attempts/notes/skills, and supports long-running multi-agent evolution.
Open-source infrastructure for continually self-improving agents. Reef serves inference requests, records feedback, produces updates, and commits accepted model-weight or agent-harness improvements to a version history.
Market Position
CORAL positions itself as an autonomous alternative to fixed evolutionary-search systems such as OpenEvolve and AlphaEvolve: agents choose what to inspect, test, remember, and communicate rather than operating inside a rigid mutation/population scaffold. It also differs from fixed-role multi-agent harnesses through horizontal parallel exploration, shared persistent memory, and agent-controlled collaboration. Reef extends the organization’s position into continual self-improvement from interaction and feedback.
Founding Story
The project grew from research into whether LLM agents could autonomously explore open-ended optimization problems rather than follow rigid, human-designed evolutionary loops. The team built CORAL to give agents autonomy while adding reliable long-running infrastructure: isolated workspaces, separated evaluation, persistent shared state, asynchronous multi-agent execution, and heartbeat-based reflection.
Target Markets
- AI research labs and academic researchers
- ML engineers and data-science teams
- Systems and GPU-kernel engineers
- Developers building autonomous coding-agent workflows
- Teams experimenting with continual learning and self-improving agents
- Autonomous code optimization and autoresearch
- Mathematical discovery and open-ended problem solving
- GPU and systems-kernel engineering
- Machine-learning engineering and benchmark optimization
- Multi-agent experiments that need persistent knowledge and long-horizon iteration
- Continual improvement of model weights or agent harnesses via Reef