Troubleshooting

Check the model-specific power requirements

Use the supply and cable specified in your device guide. THOX products have different power requirements; do not assume one wattage or connector applies to all of them.

Inspect the connection

Check the cable and power connection for damage or a loose fit. Use a known-working outlet and a compatible replacement cable if the product guide permits it.

Follow the documented startup sequence

Use the indicators and wait times given for your model. Record any error or repeated startup pattern.

Use the documented recovery route

Do not press an unidentified reset button or erase the device as a first step. Contact support with the model, software version and symptom if normal startup does not work.

Check the local connection

Confirm the supported network interface is connected and has the expected address. A guest network or managed-network policy may prevent peer traffic.

Confirm the actual destination

Use the address shown by setup or the application. Do not assume a common hostname, port or LED color across THOX devices.

Read each MeshStack state separately

Account access and a recent device heartbeat do not prove tunnel connectivity. Adapter running with Connecting means the peer handshake has not been established.

Check the platform and peer

Confirm that both peers use a supported native connection path. The browser console does not create a VPN tunnel, and the current macOS Tauri tunnel path reports UnsupportedPlatform.

Keep network changes scoped

Follow the app-specific permissions instructions and ask your network administrator about blocked routes. Do not disable UAC or the firewall, remove another VPN or open router ports to force a connection.

Identify where inference runs

Check the selected runtime or hosted provider first. Local hardware load and hosted service availability require different checks.

Check the workload

Try a short synthetic prompt and a model supported by your runtime. Long context, large outputs and concurrent work can increase response time.

Check device conditions

For local inference, inspect memory, storage, running tasks and any temperature indicators provided by the device software. Follow the model-specific ventilation guidance.

Compare a repeatable example

Record the model, runtime version, prompt length and time to response. Keep the example non-sensitive so support can understand the issue without private data.

Follow hosted retry guidance

Check the displayed error and allowance before retrying a demo. Repeated requests can consume the remaining usage limit without resolving an unavailable provider.

Check the service and credentials

Use the endpoint and access method for the installed runtime or website service. EdgeLab requires a signed-in account with a confirmed email; a device API can use different credentials.

Refresh changed provider details

In EdgeLab, a 409 provider-selection error means you must refresh the displayed provider and model and review consent again before submitting.

Respect usage limits

A 429 response means a limit was reached. Follow any retry guidance shown and wait for the allowance to reset; do not repeatedly resubmit or create sessions to evade the limit.

Check availability before retrying

For an unavailable service or a timeout, retain the error code and check the application state. A configured provider does not guarantee a successful inference request.

Share a sanitized report

Include the page or product, versions, time, error code and steps to reproduce. Exclude passwords, cookies, provider keys, private prompts and personal files.