Knowledge base
Ingest content, embed it, and let the AI agent answer from it.
A knowledge base is the content your AI agent answers from. You can create multiple knowledge bases per workspace (up to your plan's limit — 1 on Free, 3 on Starter, 10 on Growth, unlimited above). Content is broken into chunks, embedded into vectors, and searched by similarity at answer time.
Ingestion sources#
Fetch a public page and import its text.
PDF (extracted to text), Word (.docx), and plain text/HTML.
Paste content directly (stored as a text source).
Source limits & safety#
- URL crawl is protected against SSRF — only
http(s)URLs are allowed and requests to private/internal addresses are blocked. Fetches time out quickly and cap the page size, so point it at focused pages rather than huge documents. - Pasted text is capped per source (around 100,000 characters).
- A knowledge base has an overall size budget (about 10 MB of source content).
How content becomes searchable#
Chunking
Sources are split paragraph- and sentence-aware into chunks (roughly 800 characters with 200 characters of overlap) so retrieval stays coherent.
Embedding
Each chunk is embedded into a 1024-dimension vector using the platform's embedding model.
Vector search
Vectors are stored with a similarity index. At answer time, the agent retrieves the closest chunks to the customer's question.
Reindex
Re-running indexing clears existing chunks and rebuilds them from the current sources. Indexing status is shown per source (pending, success, failed) and polled until done.
Editing a document#
Open a file or pasted-text source to edit its full text in one editor, rather than patching individual chunks. The text you see is the document Chatasist actually indexes — for PDFs and Word files, that's the extracted text, so you can clean up any extraction glitches right here.
When you Save, Chatasist replaces the stored text and re-indexes just that document in the background: it rebuilds the search index from scratch, so what the AI retrieves always stays consistent with what you edited. You don't manage the internal segmentation yourself — just edit the document and save.
Editing a document is available on every plan that includes a knowledge base. URL sources and sources from connected apps (Notion, Confluence, and other integrations) are read-only here — they're refreshed from their origin each time the base is re-indexed, so edit them at the source instead.
Attention signals on the knowledge list#
The knowledge base list shows two badges that point to bases needing your attention:
- Needs attention — set when Chatasist detects conflicts across sources, or when a specific source failed to index or contains issues. Open the base to see which source triggered the warning and fix or remove it.
- Reindex due — appears on bases with automatic re-indexing enabled when the last refresh is past its expected cadence (or has never run). Use the Reindex now action from the card's menu to refresh manually.
These signals are visible at a glance on /knowledge so you can keep the AI's source
content healthy without opening each base.
Conflict detection#
Chatasist flags a knowledge base for attention when sources appear to contradict each other, so you can resolve conflicting answers before they confuse customers.
Your Help Center News and Help articles are also maintained as automatically managed knowledge so the agent can answer from them.
Let the agent read live customer data, not just static content.