Reef: Open-Source Infra for Self-Improving Agents Gains ~300 Stars in 2 Days
pliang279 · x · 2026-09-04
Reef, open-sourced by the Human-Agent-Society team, is the first open-source infrastructure that evolves both model weights and the agent harness from live experience — it earned 300 GitHub stars within two days (272 stars, 22 forks at time of writing).
- It provides a full continual-learning backend with standardized HTTP endpoints: download agents like you'd curl codex or opencode, and route model requests to Reef instead of the provider
- Core mechanism: Reef continuously evaluates agent behavior and improves the served harness and weights in the background, so results keep getting better with no user effort
- Model evolution recipes: SAO, TTT-Discover, OpenClaw-RL, with more coming
- Harness evolution: SkillClaw, Meta-Harness, GEPA, with native integrations for pi, OpenCode, and Hermes Agent
Code, docs (English and Chinese), and tutorials are on GitHub under Human-Agent-Society/reef.
More from coding & agent
- Plane rebuilds project management around AI agents, not Copilot add-ons — JosephJacks_ · 2026-09-04
- GPT-6 built a full macOS simulator in under 75 minutes, stunning the panel — thursdai_pod · 2026-09-04
- Matt Shumer reviews GPT-6 Astra: first model he trusts to run his inbox and business — mattshumer_ · 2026-09-04
- Agent refused orders for 40 minutes after mistaking Anthropic's hidden tag for prompt injection — Grimmoner · 2026-09-04
- Tardigrade: an open-source framework building agent harnesses as components over an immutable event log — viksit · 2026-09-04
- Keeping an always-on AI agent: clone a second MacBook via Migration Assistant + codex remote — hudzah · 2026-09-04