Index and search data with tools

Indexing tools aren’t limited to the Indexes tab — once a toolkit has an index, an agent, a pipeline, or a chat conversation with that toolkit attached can call its search tools directly.

The tools

Tool	What it does	Where it can run
Index Data	Creates or updates an index	Indexes tab only
Search Index	Direct semantic search over an index	Indexes tab, chat, agents, pipelines, Test tools
Stepback Search Index	Rewrites the query before searching, for better recall on complex questions	Same as Search Index
Stepback Summary Index	Runs a stepback search and returns a generated summary	Same as Search Index
Remove Index	Deletes an index	Indexes tab only
List Indexes	Lists a toolkit’s indexes	Rarely needed — the Indexes tab already lists them

Warning

Index Data can’t be tested from the Test tools pane. The Go backend routes indexing through a separate, asynchronous endpoint owned by the Indexes tab. If you press Run on Index Data from a toolkit’s Test tools pane, you get that route’s own refusal rather than a result — use the Indexes tab to actually create or update an index.

Search from chat or an agent

1. Add the indexed toolkit as a participant in a conversation, or attach it to an agent.
2. Make sure the search tool you want (Search Index, Stepback Search Index, or Stepback Summary Index) is turned on in the toolkit’s configuration.
3. Ask a question in natural language. The toolkit runs the search tool against the index and returns matching content — or, with Stepback Summary Index, a generated summary of it.

Search from the Indexes tab

You can also run these search tools directly from an index’s own page, without going through chat — useful for quickly checking what an index actually returns before wiring it into an agent.

Vector store and embeddings

Search quality depends on the embedding model indexing itself uses: it turns your query into a vector, which is compared with the indexed vectors in the platform’s vector store (Qdrant, behind the elitea-vector service). There is no vector-store configuration to set up. See Index external data for how to configure the embedding model.

Next steps

• Use the Indexes tab to create an index in the first place.
• Schedule re-indexing to keep results current.