Knowledge base

The knowledge base is a curated library of your organization's documents from SharePoint that Strata indexes and searches to ground its answers. Admins manage which SharePoint sites are indexed, monitor sync status, tune indexing limits, and control visibility through extension policies.

The knowledge base has its own admin page — open Admin → Knowledge Base. The page appears only for admins who can manage organization settings, and only when your organization's policy allows the OneDrive extension (the Knowledge Base reads SharePoint, which depends on OneDrive).

How it works

The knowledge base uses retrieval-augmented generation (RAG). You add SharePoint sites, Strata automatically downloads documents from their default document libraries, extracts text (including OCR from images and scans), chunks it, embeds each chunk using your organization's embedding model, and stores everything in the platform database. When a user asks a question in chat, Strata searches the indexed chunks by semantic similarity and includes the top matches in the AI's context — the AI then writes answers grounded in your organization's knowledge.

Indexing is asynchronous and happens in the background. A newly added site syncs immediately, then re-syncs automatically every 6 hours. Changes to documents (new, updated, or deleted files) are detected on each sync and the index stays current.

Adding a knowledge source

You must have the Manage organization settings permission, and the OneDrive extension must be enabled by your organization's policy.

1. Open Admin → Knowledge Base (if the page doesn't appear, enable the OneDrive extension under Settings → Extensions → OneDrive → Manage).
2. Click Add site.
3. Search for a SharePoint site by name in the picker. The list shows sites you have permission to see.
4. Click a site in the list to add it — there's no separate confirm step; selecting it kicks off the first sync.

You must connect OneDrive yourself first (Settings → Extensions) — the site picker runs with your Microsoft account, and so does the first sync unless the organization-wide connection is set up (see Microsoft credentials and access below).

Strata begins syncing the site's default document library immediately — this can take several minutes for large sites. The source status shows Syncing, then Indexed when complete. You can check the sync status and document count on the Knowledge Base page.

Visibility and permissions

The Knowledge Base feature respects your organization's Microsoft extension policy, which has two parts — OneDrive and SharePoint:

• The Knowledge Base administration page appears only when your organization's policy allows the OneDrive extension. If OneDrive is excluded, the page is hidden from admins.
• A user can see knowledge base results in chat only when their effective allowed extensions include SharePoint — KB content is sourced from SharePoint, so disabling that extension for a role or user turns off KB search for them even if the org has sources indexed.

SharePoint depends on OneDrive: if OneDrive isn't allowed, SharePoint can't be either, so disabling OneDrive turns off both the admin page and chat KB search.

Users see knowledge base results only when the AI uses them — there is no separate "browse the KB" interface. Results appear as source links in the chat response with document names and links back to the original SharePoint file.

Managing sources

Each source has a row-actions menu (the ⋯ button at the end of its row):

Pause or Resume — pause a source to stop searching it while keeping its indexed content; resume to search it again. (Paused sources show the Paused status.)

Sync now — re-index the site's documents immediately. Normally, Strata syncs every source automatically every 6 hours.

Reindex (full) — force a full re-extraction of every document in the source, clearing stored timestamps and byte-hashes so nothing is skipped. Use this if you upgrade the text extraction engine and want to re-capture content that was previously truncated or failed to extract. Chunks and embeddings are rebuilt on the next pass.

Remove — delete the source and all its indexed documents. The SharePoint site itself is untouched. This is permanent and cannot be undone.

File indexing limits

By default, each SharePoint site can contribute up to 2,000 documents to the knowledge base. Strata indexes files in breadth-first order (topmost folders first) until the cap is reached; files beyond the cap are ignored on every sync.

You can raise this limit on the Knowledge Base page. Next to Per-site file limit, click Edit and set a new value (or clear it to restore the default):

Increase this cap if you have legitimately large curated sites (e.g., a policy library with thousands of PDFs). Each additional file adds embedding and storage cost — only increase if necessary.

Document extraction and indexing

Strata indexes the following file types: PDF, DOCX, XLSX, PPTX, TXT, Markdown, CSV, TSV, JSON, XML, LOG, images (PNG, JPG, GIF, WebP, BMP, TIFF, HEIC), and others.

Each document is extracted up to 200,000 characters (images are OCR'd, PDFs and Office files are text-extracted). Content is split into overlapping chunks (~1,500 characters each, with ~200 characters of overlap) and embedded using the Text Embeddings model. This model is part of Strata's Model Catalog and runs through Microsoft Azure AI Foundry — your organization's own embedding deployment in its Azure AI Foundry.

When your organization has the Cohere Rerank model enabled in the Model Catalog, knowledge base search casts a wider net and reranks the candidates for sharper relevance. Reranking is an enhancement only — if it's unavailable, search falls back to vector similarity order.

Files are skipped if they:

• Exceed 10 MB in size
• Are not an indexable file type
• Contain no extractable text (e.g., binary-only files) — these are stored as empty so they aren't re-downloaded every sync
• Are unchanged since the last sync (Strata skips re-extracting and re-embedding them to save cost)

Microsoft credentials and access

Strata uses two credential paths for syncing:

1. Organization-wide Microsoft app credential (preferred) — Strata has an app-only identity that can read selected SharePoint sites once a tenant admin grants consent. This persists when the admin who added the source leaves your organization.
2. Delegated admin token (fallback) — if the org credential is not set up, syncs run with the admin's personal Microsoft account. If that admin's access is revoked, syncs fail until another admin reconnects OneDrive in their Settings → Extensions.

To set up the organization-wide credential, look for the Organization-wide access not approved yet notice on the Knowledge Base page:

1. Click Open Microsoft approval. A consent URL opens in your browser — sign in as a Microsoft 365 tenant administrator and approve the consent prompt.
2. Return to the Knowledge Base page and click Verify connection.

Once connected, the page shows Organization connection active and syncs automatically prefer the org credential. For each source, an Org access indicator shows whether Strata can currently read that site. If a site isn't reachable yet, click Grant access to give Strata read access to that specific site — this runs through your signed-in Microsoft account and doesn't require PowerShell or the SharePoint admin center (you do need to be a SharePoint administrator or an owner of the site).

If organization connection fails, syncs revert to the adding admin's delegated token. A warning message indicates when this happens.

Note

Syncs are read-only — Strata extracts document content and metadata but does not modify SharePoint.

Sync failures

Each source shows a Status in the sources table:

• Indexed — the last sync succeeded and documents are searchable.
• Syncing — currently indexing; check again in a few moments.
• Awaiting first sync — the source was just added and hasn't completed its first sync yet.
• Paused — you paused the source; it isn't searched until you resume it.
• Needs attention — the last sync failed. Hover over the status for the error detail.

Common causes:

• No Microsoft access — the org connection is not set up or the adding admin's OneDrive is not connected. Reconnect under Settings → Extensions or set up the organization connection.
• Embedding model unavailable — the Text Embeddings model is not responding. If your organization brings its own Foundry, verify the embedding deployment under Admin → AI Engine; otherwise contact Strata support.
• Site not granted — the org app does not have permission to read this SharePoint site yet. Click Grant access to fix it (you must be a SharePoint administrator or an owner of the site).
• All documents failed — a systemic issue with extraction or embedding. Retry the sync or contact support.

Individual file failures (some docs succeeded, some failed) are logged but do not block the sync — failed documents are retried on the next sync if the file has been modified.

Audit and permissions

All knowledge base changes are recorded to the audit log with the acting admin's email and source IP, and appear in the activity table under Organization Settings → Audit (see Audit log). Audited actions include:

• Adding a source
• Pausing / resuming a source
• Manual sync requested
• Reindex requested
• Removing a source
• Organization connection verified or failed
• Site access granted or grant failed
• Per-site file limit changed

Only admins with the Manage organization settings permission can manage the knowledge base.

Related

• Chat overview — how users access knowledge base results in conversation
• AI engine — the Model Catalog, including the Text Embeddings model that powers KB search
• Organization settings — enabling the Microsoft extension that the Knowledge Base depends on
• Roles and permissions — controlling who can manage the KB and who can use extensions
• Audit log — reviewing KB administration changes