Copilot Studio Friction
Import succeeded — but the agent answers nothing
Last verified
Details & related
Assessment
- Confidence
- Confirmed officialConfirmed by Microsoft documentation or an official statement.
- Severity
- Blocking
- Typical time lost
- Hours
Identification
Verification & changes
Verified
Doc check by human: Official SharePoint knowledge documentation and both practitioner posts re-read during seed migration; ALM-unsupported statement for the upload method still in place.
Change
Provisionally approved by the Product Owner; external LLM quality review pending.
Change
Initial record created from the seed dossier (migration wave 1).
Are you in the right place?
- You imported your agent as a Power Platform solution, and the import reported success.
- The agent opens without errors in the target environment.
- Questions that need your documents get no answer or a fallback reply.
- The knowledge source is still listed by name inside the imported agent.
- The same agent answers those questions fine in the source environment.
If instead the export already failed or shipped incomplete → see Solution export fails or misses components. If instead authentication, icon, or channels are missing after import → see After import: authentication, icon and channels are gone. If a knowledge URL can't be changed in production (managed solution) → (record planned). If knowledge fails in an agent you never imported → see Knowledge returns nothing — and no error is shown.
What's happening
Your knowledge traveled as a name, not as usable data. The upload method for SharePoint content builds its search index inside Dataverse, the Power Platform database. That index does not travel inside the solution, and the import does not rebuild it. ALM — application lifecycle management — means moving agents between environments. Microsoft documents openly that ALM is not supported for this knowledge method (documented as of 2026-07-08). So the import reports success, the knowledge source shows up by name, and retrieval returns nothing. Picture moving house with a library: the catalog cards arrive, but the books stay behind.
For technicians
Indexing under the upload method creates real Dataverse components: tables, records, and search configuration with semantic indexes and vector embeddings. The Dataverse search connector and the semantic index do not reconstruct themselves on import. Topics, entities, and connector references travel automatically. Knowledge sources, agent flows, prompts, connected agents, and evaluation test sets do not. The connector-based SharePoint method supports environment variables and is ALM-clean. The trade-off: it skips the semantic indexing, so retrieval quality differs (documented as of 2026-07-08).
How to fix it
Solution 1
Official workaroundReprocess the knowledge in the target environment
- Open the imported agent in the target environment.
- Open the agent's "Knowledge" page.
- Remove the knowledge source that arrived with the import.
- Add the same source again, so the target environment builds its own index.
- Wait until processing finishes.
✅ You should now see: the knowledge source shown as ready, no longer as processing.
- Ask a knowledge question in the test pane.
✅ You should now see: an answer grounded in your documents.
Solution 2
Official workaroundUse the connector method with environment variables
Use this when you deploy the same agent repeatedly across environments.
- In the source agent, remove the upload-method knowledge source.
- Add the same SharePoint site as a connector-based knowledge source.
- Create an environment variable for the SharePoint URL.
- Use that environment variable as the source URL.
- Export and import the solution as usual.
- During import, set the environment variable's current value to the target's SharePoint URL.
✅ You should now see: the imported agent answering from the target source, with no reprocessing step.
Weigh the trade-off first: the connector method skips semantic indexing, so retrieval quality can drop (documented as of 2026-07-08).
Check that it worked
Ask a question in the target environment whose answer can only come from your documents. Expected: a grounded answer, not a fallback reply. Then publish and repeat the same question in the real channel.
If it didn't work
- The index is still building. Reprocessing large sources takes time. Wait, then retest in the test pane.
- You tested in a channel too early. Propagation delay is suspect number one. See Teams answers with an old version of your agent.
- The solution is managed. Knowledge URLs can't be changed in a managed solution (record planned).
- Authentication or channels also broke in the same import. See After import: authentication, icon and channels are gone.
Prevent it next time
- Choose the knowledge method — upload or connector — as a deliberate ALM decision before you build.
- Run the ALM ritual before every export and after every import.
- Plan knowledge reprocessing time into every deployment that uses the upload method.
Evidence
Official documentationlearn.microsoft.com
Microsoft documents that ALM is not supported for the SharePoint upload knowledge method; importing an agent does not trigger knowledge source processing.
MVP blogharrytraynor.io
A practitioner deep dive documents the canonical case: the solution imports cleanly, the agent opens fine, and answers no question because the Dataverse index did not travel.
MVP blogsuparnatechbasket.wordpress.com
A Microsoft engineering blog confirms environment variable support for website and SharePoint knowledge sources, making the connector method ALM-clean.