Copilot Studio Knowledge Editor — Fabric-parity spec

Comparative positioning note

This document is written from the perspective of Microsoft Azure, Cloud Scale Analytics, and CSA Loom. Any description of third-party or competing products, services, pricing, or capabilities is derived from publicly available documentation and sources believed accurate at the time of writing, and is provided for general comparison only. We do not claim expertise in, or authority over, any non-Microsoft product or service; the respective vendor's official documentation is the authoritative source for their offerings, which may change over time. Nothing here is intended to disparage any vendor — where a competing product has genuine advantages, we aim to note them honestly. Verify all third-party details against the vendor's current official documentation before making decisions.

Captured 2026-05-26 by catalog agent from Microsoft Learn (Copilot Studio: knowledge-add-existing-copilot, knowledge-add-file-upload, knowledge-add-sharepoint, knowledge-add-dataverse, knowledge-azure-ai-search, knowledge-unstructured-data, knowledge-real-time-connectors) + inspection of apps/fiab-console/lib/editors/copilot-studio-editors.tsx::CopilotKnowledgeEditor and listKnowledgeSources / addKnowledgeSource in apps/fiab-console/lib/azure/copilot-studio-client.ts.

Overview

Knowledge sources let a Copilot Studio agent ground its generative answers in your data. Each agent can attach multiple knowledge sources of mixed type. At runtime the orchestrator (when generative mode is on) or the Conversational boosting system topic (classic mode) searches across the attached sources, retrieves the top semantic matches, and uses them to compose a grounded answer with citations. Sources can be attached at the agent level (any topic can use them) or at the node level (scoped to one Generative answers node).

Copilot Studio UX

Knowledge page chrome

• Overview · Knowledge entry points → opens Add knowledge dialog
• Knowledge list table: name · type icon · status (Ready · Indexing · Error · Updating) · last-indexed timestamp · row-level Edit / Delete / Disable
• Filter by type · Search by name · Reindex action (some types only)
• Knowledge attached at the Generative answers node level is shown inline on the node, not on the Knowledge page

Source types (Add knowledge dialog)

Type	Configuration	Backing infra	Sync
Public website	URL (up to 4 root URLs) · Bing site-search · advanced: per-URL allowlist/denylist	Bing index	Real-time crawl
File upload	Drag/drop or browse from local · .docx .pptx .pdf .xlsx (+ images for PDF OCR) · max 512 MB · max 500 objects per agent	Dataverse (chunks + vector embeddings)	Static — re-upload to refresh
OneDrive	Browse-items file picker · select files/folders · ≤5 root items	Dataverse copies + SharePoint search	Auto every 4–6 h
SharePoint (file upload variant)	Site URL + Browse picker · supports SharePoint lists + files/folders · authenticated user must have access	Dataverse copies + SharePoint search	Auto every 4–6 h
SharePoint (connector variant)	Site URL · auth: shared / per-user · advanced filters (author / modified date / title) · no Dataverse copy	Live SharePoint search	Real-time
Dataverse	Pick table from current env · select columns · grounding strategy: Single best record vs Multiple top records · per-row security honored	Dataverse natural-language search index	Real-time
Azure AI Search	Service name · index name · API key / managed identity · embedded fields config (title · content · vector)	Azure AI Search service	Index-driven
Real-time connectors	Pick from 1500+ Power Platform connectors (Salesforce / ServiceNow / Confluence / Zendesk / Jira / Snowflake / etc.) · connector connection ref · per-connector filter	Live source query through connector	Real-time
Manual entries / Q&A pairs	Pairs of question + answer; useful for FAQ overrides	Stored in msdyn_knowledgesources of type manual	Static

Per-source metadata

• Name — must be unique on the agent
• Description — used by orchestrator to decide when to query this source; verbose descriptions improve routing
• Enabled toggle — disables without deleting
• Citation behavior — show citations · suppress citations · numbered vs inline
• Scope — agent-level (default) or specific Generative answers node only

Indexing / sync status panel

• Status: Ready · Indexing · Error · Updating · Quota exceeded
• Last successful sync timestamp
• Item count (files indexed / pages crawled / rows in scope)
• Error log — per-item failure reason (e.g. password-protected, sensitivity-labeled, >512 MB)

Knowledge in Generative answers node (per-topic override)

• Topic-level Use knowledge sources picker (subset of agent-level sources)
• Disable files for this node (e.g. medical-info topic shouldn't use marketing PDFs)
• Pre-defined data sources (legacy: Bing Custom Search · Azure OpenAI on Your Data · Custom Data) — classic-mode fallback

What Loom has today

From apps/fiab-console/lib/editors/copilot-studio-editors.tsx::CopilotKnowledgeEditor and app/api/items/copilot-studio-knowledge/**: - Env picker + agent picker (reuses the agent editor's env/agent state) - List knowledge sources for the selected agent (GET /msdyn_knowledgesources?$filter=_msdyn_copilotid_value eq {agentId}) - Per-source row: name · type · URI · status - Add knowledge dialog with a type dropdown supporting url | file | sharepoint | dataverse-table, plus name + URI fields - POST to /msdyn_knowledgesources with msdyn_copilotid@odata.bind to attach to the agent - Delete per row - MessageBar surfacing Copilot-Studio-not-enabled 503 and Dataverse errors verbatim

Gaps for parity

1. Source-type coverage — Loom only supports url | file | sharepoint | dataverse-table. Missing: OneDrive, SharePoint connector variant, Azure AI Search, real-time connectors (Salesforce/ServiceNow/Confluence/Zendesk/Snowflake), manual Q&A entries
2. File upload — Loom takes a URI string; no drag-and-drop, no .docx/.pptx/.pdf/.xlsx mime gate, no 512 MB enforcement, no actual file upload to Dataverse (the file content has to land as a msdyn_file annotation/note, not just a URI)
3. OneDrive / SharePoint browse-items picker — Loom can't open the Graph item picker; the maker must paste a URL
4. Dataverse table picker — no schema browser; the maker has to type the table logical name (e.g. account) blind
5. Dataverse grounding strategy — Single best record vs Multiple top records not exposed
6. Azure AI Search wiring — no service-name / index-name / vector-field config form
7. Real-time connector picker — no list of 1500+ connectors with connection-ref selection
8. Description field — Loom doesn't write msdyn_description on the source; orchestrator routing quality suffers
9. Status surface — Loom shows raw msdyn_status string only; no Ready/Indexing/Error/Updating chip, no item count, no error-log drawer
10. Last-indexed timestamp — not displayed
11. Reindex / refresh-now button — not exposed
12. Citation behavior — show/suppress citations toggle not exposed
13. Enabled/disabled toggle — Loom only has delete; no soft-disable
14. Per-node knowledge scoping — Loom always attaches at agent level; can't scope to a single Generative answers node
15. Quota visibility — Dataverse storage consumed by knowledge files is not shown; the maker hits 500-object/512 MB walls blind
16. Sensitivity-label / password-protected file detection — not surfaced (Microsoft Learn calls out these never index but show as Ready)

Backend mapping

Dataverse Web API on the env's instance URL: - List/CRUD — msdyn_knowledgesources (id · name · type · uri · description · status · _msdyn_copilotid_value · createdon) - File payload — files live as annotation (Note) records bound to the msdyn_knowledgesource row, with documentbody (base64) + mimetype + filename; Loom needs to POST the annotation alongside the source row - Connector knowledge sources — msdyn_knowledgesources of type connector + connection-reference GUID; the connection itself is in connectionreferences table (created via the connector picker) - Azure AI Search sources — msdyn_knowledgesources of type azureaisearch + JSON config in msdyn_configuration (service URL · index name · key vault ref or API key · vector field name) - Sync trigger — bound action Microsoft.Dynamics.CRM.msdyn_RefreshKnowledgeSource (reindex now) - Status events — Dataverse cdsentitydataservices change events; Loom can poll msdyn_status and modifiedon

Required Azure resources / tenant settings

• All Agent-editor prerequisites (env + Copilot Studio enabled + Dataverse App User)
• Dataverse search enabled — Environment Settings → Product → Features → Dataverse search = On (required for file/SharePoint/OneDrive knowledge — Microsoft Learn flags this prerequisite)
• PowerAIExtensions solution version 1.01.688+ — required for SharePoint/OneDrive file-upload variant
• For Azure AI Search source: existing Azure AI Search service (Microsoft.Search/searchServices) + index built with vector field; key vault secret or managed identity for auth
• For real-time connectors: connector connection in the env (connectionreferences); per-connector licensing (premium connectors require Power Apps / Power Automate per-user license)
• For SharePoint / OneDrive: Sites.Read.All + Files.Read.All granted to the Copilot Studio SP (admin consent)
• For unstructured data licensing: messages count against Copilot generative-answers billing; storage counts against Dataverse storage entitlement

Estimated effort

3 sessions. File upload (multipart → annotation) + drag-and-drop + mime gate + 512 MB check is ~1 session. OneDrive/SharePoint Graph browse-items picker + connector picker + Dataverse table schema browser is ~1 session (Graph + connector-list API integration). Azure AI Search wiring + status chips + reindex button + description field + per-node scoping is the third session. Real-time connectors deeper integration (1500+ connectors with auth flow) can spill into a fourth session if the connector picker has to ship full UX.