pickai
Filter, score, and recommend AI models across providers.
Metadata-first, powered by the models.dev catalog. Zero dependencies.
Metadata-first, powered by the models.dev catalog. Zero dependencies.
Why pickai?
Section titled “Why pickai?”Catalogs like models.dev list thousands of models and their metadata. AI SDKs handle the integration: authentication, streaming, provider-specific APIs. But choosing the right model for a task, weighing pricing against capabilities, context size against cost, recency against knowledge freshness, is still a manual exercise. pickai is the missing layer between “here are thousands of models” and “use this one.”
- Runtime model discovery: Build apps that surface models dynamically. Model choosers, comparison tools, multi-model routing. pickai provides scored, filterable results from live data so your app doesn’t hardcode model names that go stale.
- Build-time candidate selection: Narrow hundreds of models to a handful worth evaluating. Get a ranked starting list for your use case, validate with your own tests, and ship with confidence.
Features
Section titled “Features”- Always current: Powered by the models.dev catalog, not hardcoded lists
- Scored ranking: Weighted criteria with built-in purpose profiles for common use cases
- Fully extensible: Custom scoring criteria, filters, and constraints
- Benchmark-ready: Bring your own benchmark data as custom scoring criteria. See the docs for a working example.
- Zero dependencies: No npm packages. Uses the platform
fetchAPI (Node 18+, Deno, Bun, browsers, etc.).