The Retrieval screen (titled Agent Retrieval in the CMS) controls how the agent looks things up in its knowledge base: whether it grounds answers in your content, which content it may use, and how search and reranking behave. Found in the agent sidebar under Knowledge → Retrieval.
Reference
| |
|---|
| Enable Reference | Lets the agent answer using your documents, videos and websites. |
| Restrict Answer | When on, the agent answers only from your knowledge base. When off, it may also use information from outside it. The helper text under the switch describes the current state ("Allow the agent to respond using information from outside the knowledge base" means it is off). Available only while Enable Reference is on; turning Enable Reference off also turns it off. |
Content Scope
| |
|---|
| Content Types Configuration | The content types the agent may use, for example FAQ, Policy or Documentation. When you add types, the agent only uses content of those types. Leave the list empty to allow all types. |
| Add New Content Type | Adds an entry to the list. Each entry has a delete button. |
Search
| |
|---|
| Reference per-Query | How many references are used as sources for each user question (required while Enable Reference is on). A new agent starts at 3; the 4 shown in an empty field is only a placeholder. |
| Search Tolerance | A slider from 1 to 10 that sets how closely the content must match the question. A lower level gives more accurate results but may find no answer at all. |
| Tolerance Increment | A slider from 1 to 3. The tolerance is raised step by step, by this amount, up to the Search Tolerance level. |
Reranker
| |
|---|
| Enable Reranker | Uses an AI model to reorder the retrieved results by relevance to the question. |
| Reranker AI Model | The model that reranks the results (required when the reranker is on). |
| Documents to Rerank | How many top results are reranked (required when the reranker is on). Must be greater than Reference per-Query. |
Training
| |
|---|
| Content Training AI Model | The model used for knowledge base training and embedding generation. |
Saving
| |
|---|
| Save | Saves all retrieval settings after you confirm. |
Related guide: Retrieval