Worker Model Switch

Switch a Worker's LLM model using the agt CLI. The controller handles all details: model parameter resolution, openclaw.json generation, storage push, and container recreation.

Usage

agt update worker --name <WORKER_NAME> --model <MODEL_ID>

Examples:

agt update worker --name alice --model claude-sonnet-4-6
agt update worker --name alice --model deepseek-chat

What happens

The controller automatically:

1. Updates the Worker CR's spec.model
2. Resolves model parameters (contextWindow, maxTokens, reasoning, input modalities) from its built-in model registry
3. Regenerates openclaw.json and pushes it to storage
4. Recreates the Worker container

You do NOT need to specify context window, reasoning, or any model parameters — the controller knows them.

On success

The CLI prints worker/<name> configured. The controller reconciles the change automatically. No manual restart is needed.

On failure

Report the CLI error output to the human admin. Common causes:

• Worker name does not exist
• Controller API is unreachable

Unknown models

If the admin requests a model not in the controller's built-in registry, the controller uses safe defaults (contextWindow=150000, maxTokens=128000, reasoning=true, input=["text"]). If the admin needs specific parameters for an unknown model, they should update the controller's model registry rather than overriding at the skill level.