How to Create a Knowledge Base for Your AI Assistant

A knowledge base is a collection of sources your assistant retrieves from during a conversation — useful when there’s too much reference information to fit in the prompt, but the assistant still needs to get it right. It’s the standard way to handle support, FAQ and detailed product knowledge.

Steps

1. Open the Knowledge Base menu. In the assistants menu, click Knowledge Base.
2. Create a new knowledge base and give it a descriptive name.
3. Add data sources. Supported sources are website content via URL, documents (.pdf, .doc/.docx, .xls/.xlsx, .ppt/.pptx, .csv, .txt, .html, .md, and several other formats), and custom text snippets including FAQs.
4. Link it to your assistant so it becomes available during conversations.
5. Keep it updated as your business information changes.

How it works

Once a knowledge base is linked to an assistant, the assistant automatically tries to retrieve relevant information from it before responding — you don’t need to reference the knowledge base in your prompt for this to trigger. Referencing it in the prompt doesn’t increase how much it gets used; the retrieval happens on every response generation regardless.

During creation, sources are chunked, embedded and stored in a vector database. During a conversation, the assistant uses the conversation so far to find the most relevant chunks and feeds them to the model as context.

Best practices

• Use markdown-formatted sources with clear paragraphs — easier to chunk and retrieve accurately.
• Group related information into one chunk to improve retrieval accuracy.
• Avoid ambiguous references like “it” or “this” — use specific names and dates, since the retrieved chunk may not include the surrounding context.
• Tabular data and images are harder for the system to chunk correctly and can yield weaker results — prefer clean text where possible.

Common questions

Do I need to prompt my assistant to use the knowledge base? No — it’s fed as context automatically. Referencing it in your prompt doesn’t increase usage since the AI doesn’t distinguish “knowledge base context” from any other context it’s given.

My assistant still isn’t answering correctly — what do I check? Verify the source files are accurate and in a supported format, then use the knowledge base testing environment to test and refine answers.

Does this add latency? Usually — often 100–1000ms per response, depending on the setup.

Can I see what was retrieved for a given answer? Yes, in the knowledge base testing environment, or in your conversations tab by clicking the down arrow on a message.

Anything else that helps? Setting temperature to 0.00 in the assistant’s autopilot settings makes it more deterministic, which helps it follow the knowledge base context more precisely.

Next step

With a knowledge base in place, move on to how to create your chat and voice assistant.