How do I give an AI fitness coach reliable exercise data?
Ground your AI fitness coach in a real exercise API instead of letting the model invent movements. Have the model reference exercises by slug and resolve them through WorkoutDB, so every recommendation maps to a verified exercise with correct instructions and a real demonstration video.
The hallucination problem
Large language models confidently invent exercises, mangle instructions, and fabricate media links. For a coaching product that is a real risk: users may attempt a movement that does not exist or follow unsafe cues. The fix is retrieval — grounding the model in a trusted catalog.
A grounding pattern that works
- Give the model a tool that calls GET /v1/exercises/search and returns real matches with slugs.
- Have the model build plans by referencing those slugs rather than free-text names.
- Resolve each slug with GET /v1/exercises/{idOrSlug} to render verified instructions and the demonstration video to the user.
- Optionally pass the localized content (?lang=) so the coach speaks the user's language.
Why it fits agents
WorkoutDB is plain REST with a stable envelope and an OpenAPI 3.1 spec, so it is easy to expose as an agent tool. There is also an llms.txt guide describing the endpoints for AI consumers. Start free with 5,000 requests/month with no credit card required.
Related questions
- Can an AI agent call WorkoutDB directly?
- Yes. It is a standard REST API with an OpenAPI 3.1 spec and an llms.txt guide, so it is straightforward to register as a tool/function the model can call.
- How does this prevent hallucinated exercises?
- By having the model reference real exercise slugs returned from search and resolving them through the API, every recommendation maps to a verified exercise with correct instructions and media.
Build it on WorkoutDB
~873 exercises with demo video, ranked search, and four languages — over a clean REST API. Start free: 5,000 requests/month, no credit card.