Supported document formats
Upload PDF, Word, PowerPoint, Excel, CSV, text, and markdown files, plus web pages by URL. Files are parsed, chunked at semantically sensible boundaries, and indexed with vector embeddings for retrieval.
Hybrid search by default
The agent combines keyword and vector search so brand names, SKUs, and policy phrases get picked up alongside semantic matches. Results come back ranked, with the chunk the answer cites available to the agent.
Hallucination prevention
Agents are instructed to prefer retrieved chunks over invented text. When the search returns no good match, the agent is told to say so instead of guessing — and the question is flagged as a knowledge gap after the conversation.
Live updates, no redeploy
Replace or edit a source and new conversations start using the updated content immediately. There's no publish step and no agent rebuild — the live knowledge base is always the source of truth.
Scoped per agent
A single organisation can host multiple agents with different libraries. Scope knowledge so your support bot and sales bot read the sources that match their job.
Related docs