Reef open-sourced: continual learning infra that turns agent serving into a self-improvement loop
qinzytech · x · 2026-09-02
The Human-Agent-Society team (Ao Qu, Han Zheng et al.) has open-sourced Reef, continual learning infrastructure for agents that improve from their own experience.
The idea: instead of treating inference as the end of the pipeline, Reef turns live agent interactions into a continuous learning loop — serving real applications, capturing trajectories and feedback as structured experience, and letting different learning recipes use that experience.
Notably, what evolves is the whole agent — model weights plus the harness — with updates evaluated, versioned, and safely deployed back into serving. The repo ships standardized HTTP endpoints (agents downloadable like codex/opencode), Docker deployment, tutorials, and multiple recipes.
More from coding & agent
- Matt Pocock's pre-commit hook caps agent-written files at 5,000 tokens — mattpocockuk · 2026-10-18
- Author Laurence Moroney: write your ugly first draft yourself, then let an LLM edit it — lmoroney · 2026-10-18
- Open-weight TYGER speech decision model makes 80ms turn-taking calls for full-duplex voice agents — solyarisoftware · 2026-10-18
- Devs: Opus 5.5 on medium beats Opus on max — eyishazyer · 2026-10-18
- He Stopped Telling Claude Code to Stop Making Up Timestamps—and Made It Impossible with a PreToolUse Hook — Grimmoner · 2026-10-18
- Cursor Super + Claude Pro + OpenCode: can this stack survive heavy autonomous coding? — DarioAmodeiWasTaken · 2026-10-18