FAQ & troubleshooting

Accounts

I didn’t receive a password reset email. First check how you signed up: Google/GitHub accounts have no Insight AI password — just use the same sign-in button. For password accounts, contact your workspace administrator to regain access — reset email delivery depends on how your workspace is configured. See account help.

I keep getting signed out. Sessions last 24 hours. Signing in again is expected once a day.

I can’t add a teammate to my group. They may already belong to another group — each user can only be in one group at a time. Remove them from the other group first, and make sure they’ve created an Insight AI account with the email you’re entering.

Data sources

“Invalid SMB credentials” when creating a source. The username/password was rejected by the file server. Re-check the account details with your IT team, and confirm the account has access to the server you named. The source is only saved once credentials verify.

A share shows a red ⚠ “Access denied” in the Shares tab. The connection works, but the account can’t open that specific share. Other shares still ingest normally. Ask IT to grant read access if you need that share’s content.

The Shares tab won’t load / connectivity errors. The server may be unreachable from Insight AI. Verify the IP address or hostname and port, and that the file server allows connections from the Insight AI environment.

“Ingest everything” on Google Drive says there’s nothing to ingest. Drive access is grant-based: Insight AI can only see files you’ve approved. On a fresh connection nothing is granted yet — use Choose specific files and pick them in the Drive browser, which grants access as you select.

The Google Drive file browser won’t open. The stored Drive connection may have expired or been revoked on the Google side. Go to Integrations, Disconnect Google Drive, and connect it again.

Ingestion

My ingestion is stuck on “pending” / hasn’t started. Ingestion jobs across the workspace run one at a time. If another pipeline’s run is in progress, yours waits in the queue and starts automatically.

The run failed with an error naming a specific file. Ingestion stops at the first file it can’t process. Common causes: the file is corrupt, locked, or contains no extractable text (e.g. a scanned, image-only PDF). Fix or remove the file on the share, then select Retry Ingestion.

Chat can’t find documents that were indexed before. If a run failed partway, the collection may be incomplete — each run rebuilds the collection from scratch, so re-run the pipeline to completion. Also check that nobody deleted the collection or re-pointed a pipeline.

A file on the share never shows up in answers. Check that its type is supported, it’s under 50 MB, and it isn’t hidden or a system file. Then re-run the pipeline — ingestion reflects the share as it was at the time of the last successful run.

Chat

Answers don’t seem to use my documents. Three things to check, in order:

1. A pipeline has completed successfully (status active), so a collection actually has content.
2. You can access that collection — you must be a member of the group that owns the data source. If the collection doesn’t appear in the chat’s collection picker, that’s the problem.
3. The chat’s scope: if the pill next to the message box says something like 2 of 5 collections, the chat is limited — open the picker and widen it.

Beyond that, retrieval only includes passages clearly related to your question — try phrasing the question closer to the document’s own wording. If nothing relevant is found, the model answers from general knowledge.

The underlines and sources disappeared from an old answer. Expected: source markers exist only in the session where the answer was generated. After a reload the text remains but the citation underlines and source list are gone. Ask the question again if you need to trace sources.

“Model not found” when sending a message. The selected model has been removed from the organization. Pick another model from the selector, or re-add it via Models in the dashboard — see managing models.

Responses are slow. Generation speed depends on the model — you’ll see it directly in how fast the words stream in. Check the tok/s figure in the model selector and switch to a faster model for quick questions.

The response went wrong halfway — can I stop it? Yes. While a response is generating, the send button becomes Stop generation. The partial reply is kept with a Generation stopped chip; use its Retry link for a fresh answer, or edit your question.

Models

A model download failed. Downloads retry automatically; if it still fails, select Create Model again to restart. Check that the model ID matches the Ollama library tag exactly (e.g. llama3.2, qwen2.5:14b).

A model disappeared from my chat list. Someone deleted it from the organization’s Models page, which removes it for all users. Re-pull it or choose another model.

Something else?

If your issue isn’t covered here, contact your workspace administrator, or reach the team via insightai.pro.