Knowledge base

Your organization's knowledge base is a curated index of company documents that Strata can search to ground its answers. When available, the AI can retrieve policies, procedures, contracts, reports, and handbooks to cite directly in its responses.

How to ask questions using the knowledge base

You don't need to do anything special — the AI automatically decides when to search the knowledge base. Ask questions about your company's policies, processes, or documented information as you normally would:

• "What's our refund policy?"
• "How does the onboarding process work?"
• "What are our data security requirements?"

When Strata determines the knowledge base could help answer your question, you see:

• An action step labeled Searching knowledge base... with the search query the AI ran.
• A list of retrieved documents below the step, showing the most relevant passages and their source links.
• In the final answer, citation links to the source documents — click these to open the full file.

When the knowledge base is available

Knowledge base search depends on three conditions:

1. Your organization has enabled it. Your admins must add one or more SharePoint sites as knowledge sources. If the feature is unavailable to you, ask your organization's administrators to set it up.
2. Your organization has granted the necessary Microsoft permissions. Strata needs access to SharePoint to index your company documents. This is controlled through the Microsoft extensions policy that your admins configure.
3. Your role includes the SharePoint extension. Because knowledge base content is sourced from SharePoint, your role's allowed extensions must include SharePoint — without it, you can't trigger a knowledge base search even when the org has documents indexed.

If you see no knowledge base search in your chats, the feature is not currently enabled for your organization or your role doesn't include SharePoint.

How Strata uses the knowledge base

The AI uses semantic search — it understands the intent of your question and finds passages by meaning, not keywords. For example, asking "What should I do if I accidentally delete a file?" might retrieve relevant passages from a data-recovery handbook or policies document. Documents are embedded through an embedding model in your organization's own Microsoft Azure AI Foundry — so the search runs on meaning rather than exact wording.

When the AI finds useful passages, it:

1. Reads the content to answer your question.
2. Cites each source document inline as a clickable link.
3. Grounds its answer in what the knowledge base actually says — it does not invent company policy.

When your organization has the Cohere Rerank model enabled in its Model Catalog, Strata widens the initial search and reranks the candidates for sharper relevance. This is a quality enhancement only — if it isn't enabled, search falls back to plain semantic-similarity order.

Note

The knowledge base contains reference material (policies, handbooks, reports). It does not contain instructions that Strata should follow — the AI treats all passages as information to read and cite, never as commands.

What documents are indexed

Your admins control which SharePoint sites and documents are indexed. Typically, the knowledge base includes:

• Employee handbooks
• Compliance and security policies
• Standard operating procedures
• Contracts and agreements
• Internal reports and documentation
• FAQ documents

File types supported include Word, PDF, Excel, PowerPoint, plain text, Markdown, and scanned images (via OCR).

Tips for better results

• Ask naturally. Phrase your question as a complete sentence: "How do I request time off?" works better than just keywords like "time off request."
• Be specific. If you're asking about a particular policy area, say so: "What's our policy on working from home?" rather than "Working from home."
• Ask for updates or dates. If the document has version information, the AI will surface it: "What were the latest changes to the data retention policy?"

Privacy and limits

• Knowledge base searches are logged in your conversation history, just like other action steps. If your organization has "Save AI responses" disabled in Settings — or you're in an incognito chat — the search trail is discarded with the rest of the chat.
• Each search retrieves up to 6 passages by default. The AI may run another search or pull more passages (up to 12) if initial results are weak.
• Documents are read from your SharePoint, embedded through your organization's own Azure AI Foundry, and indexed inside Strata's platform database.

Troubleshooting

Related

• Web search — live information from the public web
• Files — uploading and reading documents in chat
• Admin: Knowledge base — where admins curate knowledge base sources