AI/ML Packages
Which libraries are agents and developers actually reaching for. Curated trending across npm and PyPI.
Two free upstream sources give us the package ecosystem signal cleanly: npm's documented public downloads endpoint and pypistats.org's JSON API (which serves the public PyPI BigQuery dataset). Combined with editorial curation of ~78 AI-relevant packages, this is the “what is the agent stack actually using” view.
Categories cover the canonical agent toolchain: LLM SDKs (Anthropic, OpenAI, Google, Mistral, Cohere, Groq, Together), agent frameworks (LangChain, LangGraph, LlamaIndex, CrewAI, Mastra, Vercel AI, AutoGen, smolagents), RAG (Pinecone, Chroma, Qdrant, Weaviate), inference (Transformers, vLLM, Ollama), MCP, evals, and observability.
npm
refreshed 03:30 UTCPyPI
refreshed 03:45 UTCFree agent endpoints
/api/packages/npm/ai-trendingnpm AI/ML packages with weekly downloads. Filter:?category=&limit=./api/packages/pypi/ai-trendingPyPI AI/ML packages with daily/weekly/monthly downloads. Filter:?category=&limit=.
Frequently asked questions
- Where do the download counts come from?
- npm: api.npmjs.org/downloads — the documented public download stats endpoint, used commercially everywhere from npmtrends to libraries.io. PyPI: pypistats.org JSON API, which serves aggregates derived from the public PyPI BigQuery dataset published by the Linehaul project (Python Software Foundation). Both are clean upstream sources free to redistribute.
- How were the package lists chosen?
- Editorial. We curated ~37 npm packages and ~41 PyPI packages each grouped into seven or eight categories (LLM SDKs, agent frameworks, RAG, inference, MCP, evals, tooling, plus observability for Python). The lists prioritize libraries developers actually reach for when building agents, not raw popularity. They're hand-edited; bumps land on redeploy.
- How often are downloads refreshed?
- Daily. npm at 03:30 UTC, PyPI at 03:45 UTC. The npm endpoint reports last-week downloads; PyPI reports last-day, last-week, and last-month. The page uses the largest available window to make rankings less noisy.
- Can I use this commercially?
- Yes. Download counts are facts derived from public infrastructure (npm registry stats and the PyPI BigQuery public dataset), redistributed under the same fair-use posture they have always had. The TensorFeed snapshot is curated and ranked, with structured attribution back to the upstream source on every response.