Select a model endpoint for your Windows agent
LibrePatch connects to a model through a standard API endpoint, to a model that runs locally on your PC, or to the built-in model. You choose the provider, test the connection and decide which model fits the task.
Three ways to run a model
Cloud AI provider
A cloud AI provider connects through a standard API endpoint that you enter in the settings together with your key. LibrePatch sends your prompts to that endpoint, and the prompt text includes the text of the files the agent reads, so the provider receives it. Check the provider's privacy terms before you point the agent at code you may not share. The key stays on your PC, and secrets in tool output are redacted before they reach the model. Providers may charge separately. Larger hosted models are often chosen for harder steps, and you review every diff before you accept it.
Local model on your PC
A model running locally on your PC is reached through a local endpoint, so the prompts go to your own machine. With a local model nothing leaves the PC, which matters when the code or its licence terms are private. The context window and the speed depend on your hardware and on the model you load. Hardware sets the limits. Use the Test button to see the response time and the context window of your setup. Web search, update checks and the phone-browser connection are separate optional network features. You still read every diff yourself.
Built-in model
The built-in model is ready to use without your own endpoint. Choose it when you want to start a first chat right after installation, and switch to a cloud or local model whenever you like. Start there. The Test button works for it as for any other model and shows its response time and context window. Whichever model you pick, you read each diff before you accept it, and Ask first pauses before actions that change something. A good first task is reading a LICENSE file and writing short notes about it.
Test the connection before you start
The Test button checks the endpoint you entered and shows its response time and its context window. Test before you start. The response time tells you how long the endpoint takes to answer, and the context window tells you how much text the model can take in one turn, which matters for large files and long command output. Run the check again after you change the endpoint or the model. A slow answer or a small window is a reason to pick another model for that step. Testing does not replace reading the diff: it only checks the connection and shows two numbers.
Pick the right model for your task
A bigger model may handle harder steps, such as a change across several files or a tricky bug, while a smaller model is often enough for a typo or a small edit. The run_on_model tool sends a hard step to a bigger model, and you can split a large job into parallel sub-agents with the task tool. Quality still varies from model to model. Compare on your own tasks. Use the Test button to see the response time and the context window of each candidate, and keep routine steps on the model you already trust.
Quality varies between models, and even between versions of the same model. Read the diff before you accept any change. Ask first pauses before actions that change something, so you see the diff and can reject it, and Plan only writes a plan and runs nothing, which lets you read the steps before anything happens. Auto-run acts at once, so use it for routine work in a folder you trust. Reading the diff helps you understand each change.