Troubleshooting
Symptom to cause to fix
| Symptom | Likely cause | Fix |
|---|---|---|
| Camera shows "Local inference requires an active paid Marketplace entitlement." | Your license is Free EDGE, a sponsored trial, expired, or revoked. | Confirm your Hosted EDGE license is an active paid Marketplace purchase. See Requirements. |
| Import fails with "Private connection profiles require HTTPS..." | Camera admin access still allows plain HTTP. | Set System > Network > HTTP and HTTPS > Admin to HTTPS only, then reimport. See Connect the camera. |
| Connection test fails: "Local inference connection failed. Check the endpoint, certificate, secret, and worker status." | Gateway is stopped, unreachable on the network, the certificate does not match, or the shared secret is wrong. | Run service status. Confirm the camera and gateway machine are on the same LAN and port 9443 is reachable. Reimport the connection file if the secret or certificate may be stale. |
| Connection test passes but sample detection fails or times out | Ollama is not running, the model is not pulled, or the machine is cold-starting the model. | Confirm Ollama is running and gemma3:4b is pulled. Run the sample again; a second attempt after model load should be faster. |
| Detections are slow (well above 3 seconds warm) | Older Ollama version, shared/loaded machine, or model not kept resident. | Confirm Ollama is 0.33.3 or newer. Use a dedicated machine. Run a sample after any restart to reload the model. |
| First detection after any restart is slow | This is a cold start. Expected. | Run a sample detection immediately after any gateway or Ollama restart before relying on timing. |
| Camera disarms and shows "entitlement inactive" | Your Marketplace entitlement lapsed, was suspended, or was revoked after arming. | Restore an active paid entitlement, then rearm from camera settings. |
| Detection stops after machine sleep or restart | The gateway service was not running, or Ollama lost the loaded model. | Run service status, restart if needed, and run a sample detection to confirm recovery. |
| Model memory contention with other apps | Another process on the same machine is using the GPU/CPU or memory the model needs. | Use a dedicated host, or run ollama stop gemma3:4b when not actively detecting, then reload with a sample before relying on it again. |
| Rotation command refuses to proceed | An unexpected file exists in the state directory. | Inspect ~/.config/anava-local for unexpected entries before retrying init --rotate. See Operate. |
If nothing above resolves it
- Confirm the installed gateway version matches the version shown in your Anava portal.
- Check
gateway.logandgateway-error.login the gateway's state directory. - Confirm no firewall or VPN is blocking TCP 9443 between the camera and the gateway machine.
- Fall back to Cloud inference from camera settings if you need detection restored immediately. See Rollback.
Related: anava.ai/edge/local