Connect a Model

LayerPrism runs your theme work by sending prompts to an AI model. You pick the provider, test the connection, and review every diff before accepting changes.

Choose your model source

Cloud API provider

Send your theme prompts to a standard cloud endpoint. The agent reads your files, sends the context, and returns results. Your API key stays on the PC. The prompt leaves your machine for that provider. You control the endpoint URL.

Local PC model

With a local model your prompts stay on the PC.

Built-in model

Use the default model included with LayerPrism. It works out of the box. It needs no endpoint of your own. Its location is not documented here, so treat prompts like those sent to a provider. You do not need extra configuration. The agent uses this by default. It simplifies setup for beginners. You can switch later if needed.

Test your connection

Press the Test button to verify your API endpoint. The tool shows the response time and context window size. This confirms the model is reachable. You see the latency clearly. You can adjust settings before starting work.

Pick the right model size

Use a bigger model for complex refactors or logic. Send hard steps with run_on_model. Smaller models handle small edits. You control the orchestration. Hard steps can be sent to a bigger model with run_on_model and the user sees which model was used.

Quality varies between models. Always review the diff. Use Ask first to pause. You see the change before it happens. You can reject bad edits. You stay in control of your theme files.