Troubleshooting
Find what you are seeing in the left column.
Signing in and sessions🔗
| What you see | What it means | What to do |
|---|---|---|
| Pages stop loading or error after Scirix has been open a while, especially after your computer slept | Your session has expired and the app has not recovered cleanly. | Sign out and sign back in. This clears it immediately. |
| A page hangs or shows "Request timed out while contacting the Scirix API" | A temporary API slowdown or connection drop. | Reload the page and retry. If it persists across reloads for more than a few minutes, report it. |
| Sign-in fails with a plain "Failed" | Your account or invite does not exist yet, or the invite was made for a different email address. | Confirm with your organization admin exactly which email the invite was created for, and sign in with that Google account. |
| A "You're already onboarded" card blocks you from the workspace | Known navigation dead-end after onboarding completes. | Sign out and back in, then open your workspace. |
| The app misbehaves on your phone | Mobile browsers are not yet fully supported. | Use a desktop browser. |
Indexing and tasks🔗
| What you see | What it means | What to do |
|---|---|---|
| Onboarding finished but nothing happens: no indexing progress, tasks never start | Your namespace or agent pool is still being provisioned, or provisioning has failed. | Wait a few minutes, then check the health surfaces. If nothing changes, report it. This is not something you can fix from your side. |
| Tasks and goals sit "queued" and nothing starts | A doc-code audit is running. Task processing is paused across all agents while it runs, and a banner says so. | Wait for the audit to finish. Queued work resumes automatically. If nothing moves long after the audit completes, check the health surfaces. |
| Tasks start but fail with model errors | The assigned model cannot actually do what its capabilities claim. Most often tool_use is checked on a model without real tool-calling support. | Uncheck the capability or switch that role to a model that genuinely supports it, under the namespace settings, LLM Assignments. |
Providers and models🔗
These rows apply when you brought your own API key or hardware.
| What you see | What it means | What to do |
|---|---|---|
| A role never uses your provider | The provider is missing a capability that role requires, is disabled, or is not in a group assigned to that role. | Check Settings, then LLM providers: capabilities, enabled state and group membership. |
| You cannot find the LLM providers page in your workspace settings | Providers are registered per organization, not per workspace. Workspace settings only have LLM Assignments, which picks between providers that already exist. | Open the user menu in the top-right corner, then Settings, then LLM providers. |
| The LLM providers page says you do not have permission, or a provider you add never saves | Managing providers needs the Owner role in your organization. Members can only view assignments. | Ask an owner to switch your role to Owner under workspace Settings, then Members & access, then sign out and back in. |
| A provider exists but a workspace cannot pick it under LLM Assignments | The provider was limited to specific namespaces when it was created, is disabled, or lacks the capabilities that role requires. | Open the provider under Settings, then LLM providers and either widen it to organization-wide or add the workspace to its namespace list. |
| "An API key is required" or authentication errors from your provider | The key is missing, mistyped or revoked. Keys are stored encrypted and never redisplayed, so a wrong key cannot be inspected, only replaced. | Re-enter the key in Settings, then LLM providers. |
Learning or an audit fails with HTTP 400 … Invalid model name passed in model=<provider> | The provider's model name has a wrong or missing route prefix. Bare names are auto-prefixed on save nowadays, so this usually means the provider was created before that, or an explicit route was mistyped. | Edit the provider's Model field. The correct form is ollama_chat/<model> for chat and ollama/<model> for embedding. Re-saving the provider re-normalizes it. If you cannot edit the provider yourself, your organization admin can. |
| Your GPU shows no activity even though Scirix is "working" | The current stage is running on a platform model, most commonly embedding during indexing. | Check the namespace settings, LLM Assignments, to see which roles your provider actually serves. |
Tailnets and endpoints🔗
These rows apply when you brought your own hardware.
| What you see | What it means | What to do |
|---|---|---|
Endpoint sync status shows failed | The reconciler could not apply your endpoint. The error snippet on the endpoint says why. | Fix what the snippet points at (address, port, auth key validity), then re-check. Verify the machine is online in your Tailscale admin console. |
Endpoint stuck in pending for more than a few minutes | The reconciler has not picked the change up. | Report it with the tailnet and endpoint name. |
When you report a problem🔗
Three details make the difference between a fixable report and a mystery:
- The page you were on. Copy the URL.
- What you did right before.
- The approximate time it happened.
Scirix logs errors centrally. With those three details, the team can find the exact failure.