Connect a Model for Quantization

ModelTrim connects to your chosen AI provider so the model can plan and write quantization scripts while Shell runs them on your PC or server. In Ask first mode, you approve every action that changes your files.

Choose Your Model Source

Cloud AI Provider

Use a cloud AI provider through a standard API endpoint. Your prompts and file text leave your PC, so sensitive data goes to that provider. This option suits complex logic where you need a stronger model than the built-in one.

Local PC Model

Run a model locally on your PC. With a local model, prompts and file text are not sent to a provider, though optional network features like web search or update checks may still use the network. This option suits sensitive work where you prefer full control over the environment. You manage the hardware resources required for inference and script execution.

Built-in Model

Use the built-in model for quick tasks without configuring external connections. This option is ready to use without your own endpoint.

Verify Connection Quality

The Test button checks your endpoint connection. It shows the response time and context window size. This helps you confirm the model is ready for quantization tasks. You see real metrics before starting any work.

Select Model for Tasks

Use a bigger model for complex refactors and logic. Use a smaller model for small edits. Send hard steps to a bigger model with run_on_model. This ensures quality for difficult quantization decisions.

Quality varies by model and task. Always review the diff in Ask first mode. This lets you see changes before they apply. You stay in control of every file modification.