Choose the AI model behind your Bash work

TechRetriever works with three kinds of model: a cloud AI provider through a standard API endpoint, a model running locally on your PC, and the built-in model. Choose before you start. Your choice decides where prompt text goes, so settle it before you open a private script.

Three ways to connect a model

A cloud AI provider

A cloud model suits drafting longer scripts and explaining unfamiliar code, and you connect it through a standard API endpoint with your own key. Mind the trade-off. The prompt text, including the text of any file the agent reads, goes to that provider, while the key itself stays on your PC.

A model on your PC

A local model suits private scripts and logs you would rather not send to a provider. Nothing leaves the PC, because the prompts never go to an outside endpoint. Quality varies with the model you load, so test it on a small script before you rely on it for a longer one.

The built-in model

The built-in model is the third kind you can choose, and there is no endpoint of your own to enter. It is one of three. The model receives the text of your prompts and of the files the agent reads, so keep passwords and keys out of your scripts; secrets in tool output are redacted before they reach the model.

What the Test button shows

Press Test after you enter the endpoint. The button checks that the endpoint answers, shows the response time and shows the context window, which is how much text the model can take in one request. A long script plus its log needs a larger window than a one-line fix.

Choose a model for the job

Use a bigger model for refactors and complex logic, such as a script with many functions or loops that depend on each other. A smaller model is enough for small edits like adding a usage message. When one step is hard, run_on_model sends just that step to a bigger model while the rest stays on your usual one.

Quality varies between models, so treat every result as a draft. Ask first pauses before actions that change something, edit_file shows a diff, and Shell reports the real exit code. Read all three before you accept.