How it works
1
Create folders and prepare documents
Organize reference material into knowledge folders, called groups in the API. Choose files appropriate for the persona’s audience, and keep facts with their conditions, exceptions, and sources.
2
Upload and wait for processing
Upload the files to a folder. Anam processes them into searchable passages. A document must reach
READY before it can be searched; check processing status rather than relying on a fixed waiting time.3
Connect a knowledge tool
Choose the folders a knowledge tool can search and attach it to your persona. Its description tells the language model when to use it.
4
Search during a conversation
When the persona calls the tool, it receives relevant passages from the selected folders and uses them to compose an answer. Test both the retrieved evidence and the answer before relying on the result.
Quick links
Prepare knowledge for your audience
A retrieved passage may appear without its surrounding pages. Give facts explicit subjects, such as the product, version, or region, and keep exceptions beside the claims they qualify. See Prepare documents for examples. Knowledge tools search the folders you attach. Check all documents in those folders for the intended audience. Labels such as “internal only” inside a file do not restrict access; see Set the knowledge scope. Keep personality and behavior instructions in the persona’s prompt. Use knowledge documents for reference facts, policies, and procedures.Access and retention
Knowledge groups and documents belong to the organization that created them. Every API request is scoped to your organization, so a folder or document ID from one organization cannot be read by another. Within your organization, a persona can search whichever folders its knowledge tools select — see Set the knowledge scope for choosing those deliberately. The knowledge endpoints authenticate with your API key, which grants access to everything your organization owns, so call them only from your server and keep the key there. Browser code never calls these endpoints: it uses a session token, which authenticates a conversation with a persona rather than access to your documents. See Authentication.Troubleshooting
Next steps
Set up your knowledge base
Create folders and choose the reference material your persona needs.
RAG cookbook tutorial
Follow a worked example of adding a knowledge base to an avatar agent.