Tool search with embeddings
Scales tool use to large tool libraries by embedding tool descriptions and selecting tools by semantic similarity before calling them.
- Upstream notebook:
tool_use/tool_search_with_embeddings.ipynb
What to focus on
- Building a tool catalog (names + descriptions)
- Embedding + nearest-neighbor retrieval to shortlist tools
- Keeping a deterministic execution layer after the model selects tools
Run locally
make test-notebooks NOTEBOOK=tool_use/tool_search_with_embeddings.ipynb