Skip to main content
A knowledge base stores your documents as embeddings so an agent can retrieve relevant context and ground its answers in your data (retrieval-augmented generation).

Creating a knowledge base

Give it a name and description, and optionally tune how strict the retrieval matching should be — stricter settings favor precision, looser settings retrieve more broadly related content.

Adding sources

A source is one unit of content inside a knowledge base — plain text, an uploaded file, a URL, Google Drive doc, or API-fed content. You can upload files directly (PDF, DOCX, TXT, MD, CSV, HTML, JSON — up to 50MB) or add a source by URL.

Ingestion

Adding a source doesn’t make it searchable by itself — it must be ingested (chunked and embedded) first. You can re-ingest a single source or every active source in the knowledge base at once.

Attaching to an agent

Link a knowledge base to an agent when creating or updating it to enable retrieval-augmented responses grounded in that knowledge base.
Keep each source focused on one topic — a single 200-page manual retrieves worse than the same content split into a handful of topic-scoped sources.