Skip to Content

Reading runtime failures

The alpha should show real blockers instead of covering them with success copy.

Failure and blocker language

Use the blocker shown by the runtime. Common classes include:

  • model provider unavailable
  • model call timeout
  • runtime fetch boundary
  • evidence unavailable
  • preview unavailable
  • worker timeout
  • local prerequisite missing

First checks

  1. Confirm the model endpoint is reachable.
  2. Confirm the model id is tarx-computer.
  3. Confirm the prepared app-support frontend and backend are running.
  4. Check the run detail endpoint.
  5. Check events, review, files, terminal, and preview for the same run id.

Model lane truth

The current local lane is NVIDIA Nemotron GGUF through llama.cpp. It is not NIM or TensorRT.

Safe support wording:

The alpha is using the local tarx-computer model lane. This run stopped on a runtime blocker, so inspect the run evidence before retrying.

Avoid:

TARX cloud failed.

Avoid:

Supercomputer capacity was unavailable.

Those claims are not part of the current alpha proof.

Next

Continue to Known limitations and claim locks.

Agents