A library wants staff research agents to call the same retrieval pipeline used for the public catalog. How should they expose that pipeline?

A library that wants staff research agents to use the same retrieval pipeline as the public catalog should expose that pipeline by connecting it directly as an agent tool. Tool-calling keeps research answers grounded in the live catalog index, applies the same ranking and filters as public search, and avoids divergent copies of holdings data. Connecting the retrieval pipeline as a fine-tuned completion model for staff research replaces live lookup with generated text and does not reuse the production retrieval pipeline with its live ranking and filters. Connecting the retrieval pipeline via nightly index snapshots for staff research leaves staff working from stale copies that drift from the live catalog and bypasses interactive calls to the live index. Connecting the retrieval pipeline as an offline scoring harness for staff research prevents interactive tool calls for live holdings questions and confines the pipeline to batch evaluation instead of grounded research. Exam caveat: agent grounding to an existing municipal retrieval system is done by wiring that pipeline as a tool—not by connecting the retrieval pipeline as a fine-tuned completion model for staff research or connecting the retrieval pipeline via nightly index snapshots for staff research. Operational check: register the catalog retrieval API or index query as an agent tool with staff auth scopes; test research prompts that must return call numbers and availability from the live index; confirm the agent never answers holdings questions without a tool call.