Reef
Open-source infrastructure for continually self-improving AI agents
Visit reefinfra.ai ↗What is Reef?
Reef is an open-source platform that enables AI agents to learn and improve continuously from live interactions. It connects agent inference, user feedback, model training, and versioned deployment in a unified system. Users can train model weights using frameworks like Slime and SGLang, or optimize an agent's harness including prompts, rules, and skills.
Designed for developers building agents that need to evolve over time, Reef supports multiple learning approaches: model weight training for personalized performance, harness optimization without local GPUs, and test-time training for scientific discovery. The project is built with Claude and released under Apache 2.0.
Momentum
Daily activity over the last 30 days: upvotes, unique visits, new GitHub stars and Hacker News points.
How do we know it's vibe coded?
vibetrend discovered Reef automatically. Evidence: GitHub Human-Agent-Society/reef; repo has AGENTS.md, CLAUDE.md. View source.
Similar vibe-coded apps
#1 this weekOpen-source local voice cloning, dubbing, transcription, and audiobook creation
#2 this week
#5 this weekSlash commands and agents for engineers using AI without compromising standards
#6 this week523 free lessons building AI systems from math and first principles
#8 this week
#10 this weekAI-powered design intelligence toolkit for building professional interfaces
Manage teams of AI agents to run your business from one dashboard
Persistent memory for AI agents across Claude Code, Cursor, and MCP clients