Search
The Search tab is a playground for testing your Content Store. Enter a query and see the chunks the store returns — the same retrieval an agent performs at run time — so you can confirm the store is useful before wiring it into an agent.
By the end of this page you will be able to run a test search and tune how results are retrieved.
Prerequisites
- Admin or Developer role.
- A Content Store with indexed content — see Add documents or the Web crawler.
Run a test search
Enter a query
On the Search tab, type a natural-language query into the Search Playground box and search.
Read the results
Each result shows its similarity score, the source (document or web page), and the matching chunk. Open View document to see the source. A metrics panel reports embedding and search latency and how many chunks were scanned.
Tune retrieval
Expand Advanced Options to adjust how much comes back:
- Results (top_k) — how many chunks to return.
- Min. Similarity Threshold — the minimum similarity a chunk needs to be included. Raise it for stricter, more precise matches; lower it to return more.
If searches return too little, lower the threshold or raise top_k. If they return loosely related chunks, raise the threshold. These are the same levers that shape what an agent retrieves.
Common issues
- No results — the store may have no indexed content yet, or the similarity threshold is too high. Add content, or lower the threshold.
- Irrelevant results — raise the threshold, or reconsider the store’s chunking settings (which are set at creation).
Next
- Give an AI Agent memory — use this store for an agent’s knowledge and memory.
- RAG tools — how agents search a store in a workflow.