Human-Agent-Society/reef
Continual learning infra for self-improving agents
About the project
Continual learning infrastructure for self-improving agents: connects agent inference, feedback collection, learning, and versioned delivery of updates. Lets you train model weights or improve an agent's harness (prompts, rules, skills).
Useful for
- Run Reef as an OpenAI-compatible inference server for an agent
- Build a weight-training loop from response scores via /reef/report
- Improve agent prompts and rules without local training GPUs
README summarized by DeepSeek V4.1 Flash. Details may be inaccurate.
Why it’s trending
- 58 stars today.
- The repository is 13 days old and already has 1,218 stars.
- Top new repositories this month: #53.
- About 181 forks a day — people are taking the code.
Stars per day
Bars are daily stars, the line is the usual pace. Red marks spike days.
Numbers
- Total stars
- 1,218
- Stars in a day
- 59
- Forks
- 93
- Issues and pull requests
- 443
- Watchers
- 29
- Language
- Python
- License
- Apache-2.0
- Latest release
- v0.0.2 · September 2, 2026
- Created
- August 31, 2026
- Last push
- September 13, 2026
Star trust
Growth looks organic: forks and discussion are in line with active projects, and stars arrive unevenly, the way people give them.
These are heuristics, not a verdict: we judge by the repository’s behavior, not by a list of stargazers.
Spotted in
- September 13, 2026Top new repositories this month: #53
Share
README badge
Paste it into your README — the badge shows the current magnitude and links to this page.
[](https://gitnova.dev/en/r/Human-Agent-Society/reef)Similar projects
-
6.8
melgarafael/DeskcommCRM
Self-hosted CRM with native AI agents for selling over WhatsApp: agents reply, qualify leads and close sales while your data stays on your server. An open alternative to Kommo, Octadesk and Intercom.
-
6.2
kruzovic7/ai-data-extractor
Python tool that extracts local chat histories from AI coding assistants (Claude Code, Cursor, Windsurf, Aider, Cline/Roo Code and more) into a single normalized JSONL format.
-
6.2
ayghri/i-have-adhd
A skill for a coding agent (Claude Code) that makes it lead with the answer and numbered steps, without preamble or filler. Helps you get concrete actions instead of long-winded reasoning.
-
6.1
tech-leads-club/agent-skills
A registry of validated skills (instruction plugins) for AI coding agents: extends Claude Code, Cursor, Copilot and others with secure, tested capabilities.
-
6.1
jordan-gibbs/hyperresearch
A deep research harness for Claude Code: turns one prompt into a report with verified citations and a persistent source vault. Every source lands in a searchable store reused by later sessions.
-
6.0
NVlabs/SoL-Pi
An extension for Pi that cuts token usage and redundant turns of coding agents via four mechanisms: action fusion, observation packing, log reduction, and context compaction.