How to use GPT0: your first local AI conversation

Choose a model that fits your device

Open [GPT0](/) and find the model selector in the welcome screen. The lightweight option is a sensible first attempt on a phone or a machine with limited memory. The balanced option is a useful starting point on a capable computer. Neither choice guarantees compatibility.

Read the download and GPU-memory estimates shown beside the selector. A model that fits on disk can still require more available memory while it runs. Close unnecessary demanding apps before the first load, and use a connection where a large download is acceptable.

Load the model before sending a prompt

Select Load model and keep the page open while it prepares the runtime and model files. The progress display describes the load; it is not an estimate of answer quality. You can cancel a download if you selected a model that is too large.

The message box becomes usable when the model is ready. If loading fails, read the error and try the lightweight option before repeating a large download. The loading troubleshooting guide explains how to separate network, storage, and compatibility problems.

Give it one clear job

For the first conversation, avoid personal documents or a complicated multi-part project. Ask for an explanation, a short outline, or an edit of a harmless paragraph. State the audience and the output format so you can judge whether the reply followed your request.

• Read the answer for accuracy rather than assuming fluent text is correct.
• Use a follow-up to correct a specific problem, such as an omitted constraint.
• Use Stop if a reply is going in an unhelpful direction.

Explain what a browser tab is to someone who is new to computers. Use one everyday analogy and three short sentences. Avoid technical jargon.

Keep useful work and know the limits

A conversation is saved in the current browser when local storage is available. Export chat downloads a Markdown copy you can keep outside the site. New chat begins a separate conversation without turning it into a different model.

GPT0 does not provide live web search, image generation, or code execution. Text-file attachments are read locally, but a long file can exceed the model's usable context. Start small, check the result, and move to a larger model only when your device can support it and the task benefits from the change.