Use semantic retrieval to select tools from a large library

Goal

Avoid loading hundreds of tool definitions into the model context at once.

Steps

1. Prepare tool name, description, parameter JSON, and permission labels; index a readable representation. [tool-search]
2. Follow the notebook with SentenceTransformer("all-MiniLM-L6-v2") for local embeddings and semantic top-k lookup. [tool-search]
3. Pass candidates only to a permission-controlled executor; measure Precision@k and missed tools on a labeled set. [tool-search]

How to verify

Even with a large catalog, context contains only search and a few candidates; low-confidence results do not execute.

Troubleshooting

If semantic search misses exact tools, add name/keyword lookup or raise top-k, then remeasure precision.