Knowledge Retrieval
The Knowledge Retrieval template searches an existing Knowledge Base and returns the most relevant results for a given query. It is a standalone retrieval component — you can use it on its own to inspect what's in your Knowledge Base, or connect it to an LLM to build a full RAG (retrieval-augmented generation) pipeline.
How the flow works
_10Text Input → Knowledge Retrieval → Chat Output
The Text Input component accepts a search query. The Knowledge Retrieval component performs a semantic search against the selected Knowledge Base and returns the top matching chunks. The results are displayed in Chat Output.
Prerequisites
- A populated Knowledge Base (use the Knowledge Ingestion template to create one)
Quickstart
-
In Orbis Studio, open the Templates library and select Knowledge Retrieval.
-
In the Knowledge Retrieval component, select your Knowledge Base from the dropdown. If it doesn't appear, click Refresh List.
-
Open the Playground, type a search query, and press Send. The matching chunks from your Knowledge Base are returned as the response.
Key components
| Component | Role |
|---|---|
| Text Input | Accepts the search query |
| Knowledge Retrieval | Searches the Knowledge Base and returns the top-matching chunks |
| Chat Output | Displays the retrieved results |
Configuration tips
- Number of results: By default, 5 results are returned. To change this, click Controls at the top of the Knowledge Retrieval component and adjust the result count.
- Search query override: You can hard-code a query directly in the Knowledge Retrieval component's Search Query field and run the flow without the Text Input component — useful for testing.
- Connecting to an LLM: To turn retrieval into a full Q&A flow, add a Prompt and Language Model component downstream. Pass the retrieved chunks into the prompt as context, and connect the user's question from Chat Input.
Next steps
- Combine with Knowledge Ingestion to build an end-to-end pipeline that ingests and queries your own content.
- Add a Language Model and Prompt to transform raw retrieval results into natural-language answers.