1 comment

[ 4.4 ms ] story [ 19.0 ms ] thread
We built a small tool that makes API access for LLMs boring and predictable.

There is a skills repository describing APIs (schemas, auth, constraints); any model can download skills it needs;

A tiny binary executes the actual API calls based on those skills.

If an API has a skill, the model can use it — same interface, same flow. Works equally well for local models and hosted ones.