How do I get a bot to answer from my own documents and show where each answer came from?

In Botlit[1], grounding an assistant in your own documents is handled through Knowledge Bases and Retrieval-Augmented Generation (RAG):

1. Ingest and Chunk Documents: Upload your source material (such as policies, guides, or handbooks). Botlit chunks documents using configurable strategies—including recursive, semantic, markdown-aware, or code-aware splitting—and embeds them into a dedicated vector collection.
2. Link Knowledge Bases to an Agent: Attach ready knowledge bases to your agent. On each turn, the runtime embeds the user query, retrieves the most relevant passage chunks using semantic search, and injects them into the prompt.
3. Enforce Grounded Answers: Agents receive explicit instructions to answer strictly from the retrieved context and say so honestly—declining with a "No matching passages" notice—if the documentation does not contain the answer.

Under the Grounded Answers with Cited Sources[2] concept, answers display numbered source chips referencing the exact documents used, alongside a retrieval panel detailing injected chunks, latency, and similarity scores.

Concept illustration(opens in a new tab) Illustration of the concept, not a screenshot of the actual product — Grounded Answers with Cited Sources[2]

While prompt-level grounding is built into the execution layer, surfacing live-turn citations in the UI arrives with the agent runtime upgrade (detailed on the Answers Grounded in Your Knowledge[3] use-case page).

To explore how agents and knowledge bases work together, visit Botlit Features[4] or review the Botlit Use Cases[5].