Retrieval Tool
A tool the agent calls to pull relevant documents or facts from a knowledge base when it needs them.
Think of it like
A research assistant you can ask “grab me everything we have on X” and who returns the right folder.
Example
Before answering a policy question, the agent calls its retrieval tool, which returns the three most relevant handbook sections.
How it actually works
A retrieval tool wraps a search over a document store — usually vector-based — so the agent can fetch grounding on demand rather than always stuffing everything into context. Making it a tool, rather than an automatic front-end step, lets the agent decide when it actually needs to look something up. Quality still lives and dies on retrieval: fetch the wrong chunks and the agent confidently answers from them.
For product teams
How agents stay grounded in your real documents instead of vibing from memory.
For engineers
A tool exposing similarity search over an embedded corpus, invoked on demand to ground responses.
Related
- RAG — The pattern it operationalizes as a tool.
- Embeddings — How documents are indexed for it.
- Search Tool — A cousin aimed at the open web instead.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome