About PT-Edge
PT-Edge tracks 220,000+ AI repositories across GitHub, PyPI, npm, Docker Hub, and HuggingFace. Every day, it scores each project on four dimensions — maintenance, adoption, maturity, and community — producing a composite quality score from 0 to 100.
The results are published as this directory: 165,000+ pages across 17 AI domains, from MCP servers and agent frameworks to voice AI, diffusion models, and vector databases. Every page is regenerated daily from current data.
The goal is to be the canonical reference for AI infrastructure decisions. When an AI agent or a developer needs to choose between tools, this directory provides scored, structured, daily-updated recommendations — so the work of evaluating quality doesn't have to be done from scratch every time.
Data sources
- GitHub API — stars, forks, commits, contributors, license, topics, push dates
- PyPI and npm — monthly download counts, package metadata
- Docker Hub and HuggingFace — model and dataset tracking
- Hacker News and community forums — discussion signals
How categories are discovered
Categories are not hand-curated. Each project's description, README summary, and topics are embedded as a 1536-dimensional vector. These embeddings are clustered using UMAP dimensionality reduction and HDBSCAN density-based clustering. An LLM labels each cluster based on its most representative projects. The result: 2,400 search-intent-aligned categories discovered from the data itself.
Built by
PT-Edge is built by Graham Rowe at Phase Transitions — a weekly newsletter on building with AI, from architecture decisions to production patterns.