Model settings

Managed interpretation

After signing in, the CLI can use your TrialONE plan for model interpretation. Availability depends on your account and incident allowance. The terminal names the processing mode and reports when interpretation cannot run. Detection still runs without a model.

Bring a provider key

Run trialone to open the interactive CLI, then use /key to configure a supported provider and /model to choose its model. The key is saved on this machine. Requests use that provider and its billing, with the context described in Data and model processing.

Use your own endpoint

TrialONE can send model requests to an endpoint compatible with the chat-completions protocol. Your team runs the server and chooses a model that supports the tools needed for investigation. Set these values in the CLI environment or ~/.trialone/env:

TRIALONE_LLM_PROVIDER=lokalTRIALONE_LLM_BASE_URL=http://localhost:8000/v1TRIALONE_LLM_MODEL=<model-served-by-your-endpoint>

If your server requires authentication, also set TRIALONE_LLM_API_KEY. Replace the example address and model with your actual server. Explicitly choosing lokal keeps model requests on that route; a slash in your model's name does not select an external provider.

The endpoint controls where model inference happens. Use --no-push separately when you want to skip the workspace upload. The CLI still needs approved access and periodically checks it; this setup is not a complete offline deployment.

Skip interpretation for one diagnosis

$ trialone diagnose ./recording.mcap --no-explain --no-push

This runs detection and writes the report locally, without model interpretation or posting the run to the workspace. It does not disable account access checks.