Build your knowledge base
The knowledge base holds your own content: policies, product details, FAQs and help pages.
- Plan
- Depends on your plan · Which plan do I need?
- Who can use this
- Account owner, Manager
The knowledge base holds your own content: policies, product details, FAQs and help pages.
The knowledge base holds your own content: policies, product details, FAQs and help pages. Your AI agents use it to answer customers. You group content into collections. Each document you add is split into short question-and-answer chunks that AI can search.
Sidebar → AI → Knowledge Base. The page is titled Knowledge Base: "Organize, manage, and search knowledge that powers your AI agents."
You need the permission to manage chatbots (it also controls the AI menu).
ai_agent).Tip: use one collection per topic (for example products, policies, shipping). An agent can be set to search only the collections it needs.
When it is done, you see "Document ready".
In the collection's Documents tab, the Status column shows Ready, Pending Review or Error. Use the status filter to find documents that need you.
If your workspace used the older knowledge sources, open AI → Legacy Sources. Open a source and click Migrate to Knowledge Base, then Migrate now. This copies its content into a new collection and leaves the original untouched. When you have checked the new collection, you can Delete this source. Any bot that still reads the old source stops working after you delete it.
ai_agent). Check it under Billing → Current Plan..doc, but only .pdf, .docx and .txt are read in the panel. Save a .doc file as .docx first.Do it with Copilot
(a list of questions and answers).
Do it with Copilot
Do it with Copilot
Do it with Copilot
Adding, replacing or deleting sources asks you to confirm. In ask mode it needs approval in the panel. Creating a collection does not.
| What you see | Why | What to do |
|---|---|---|
| Setup steps on the AI overview page are greyed out | OpenAI is not connected | Connect OpenAI in Integrations |
| OpenAI key needed when you add, re-fetch or edit content | No OpenAI key is saved. The knowledge base needs it for search even with a Claude model | Click Add key and connect OpenAI |
| "DOCX parsing failed: … Try pasting the text manually." | The Word file could not be read | Use Paste Text |
| "Unsupported file type: …" | The file is .doc or another type | Save it as .docx, .pdf or .txt |
| "PDF parsing failed: … Try pasting the text manually." | The PDF has no readable text, or it is damaged | Copy the text and use Paste Text |
| "Ingest failed … Attempt X of 2" | Splitting or embedding failed | Click Retry. If it fails again, check your OpenAI credit |
| Status Error | Embedding ran out of retries | Row menu → Retry ingest |
| Stuck in Pending Review for over a day | Chunks are waiting for approval, or processing stopped | Approve the chunks, or re-fetch the document |
| A URL document has little or no text | The page is built with JavaScript, or it needs a sign-in | Use Paste Text, or add authentication |
| "Pinecone index unreachable" on a legacy source | The old index no longer exists | You can delete the record, but you cannot migrate it |