How we ingest your help center, product docs, policies, and pricing into a vector index your chat agent can search at conversation time. Part of the Chat agents section of the TrainYourAgent documentation, written to be executed against rather than skimmed.
When a chat agent answers a question about your product, your policy, or your pricing, the words it uses are not coming out of the language model's pre-training. They are being looked up from a retrieval index we build out of your own content. That index is the knowledge base. The quality of your chat agent is bounded by the quality of what is in that index. Garbage in, hallucinations out.
We can ingest almost anything text-shaped. Help center articles, product pages, FAQs, policy PDFs, pricing pages, terms of service, internal runbooks, past support tickets, training documents, and Notion or Confluence exports. We can also crawl your public website on a schedule so that updates flow into the index automatically. For private content, we accept exports as zip, markdown, or a CSV with one row per article. If you do not have a written knowledge base…
Each piece of content is split into chunks of roughly two hundred to four hundred words, each chunk is converted to a vector embedding, and the vector is stored in our index along with the source URL and a few metadata tags. At conversation time, the agent's query is also converted to a vector, the nearest matching chunks are retrieved, and those chunks are inserted into the model's context as grounding for its answer. Chunk…
Every chunk gets a small set of metadata tags — content type, last updated date, audience, product line. These tags let the agent filter retrieval at conversation time. A logged-in customer on the pro plan asking a pricing question should only see pro-plan pricing chunks, not the enterprise tier. A pre-purchase visitor asking about features should not be pulling from internal runbooks. Setting up the metadata correctly during ingestion saves enormous prompt-engineering effort later.
This article is part of the Chat agents section. How we ingest your help center, product docs, policies, and pricing into a vector index your chat agent can search at conversation time.