Point your editor at a local model
The engine exposes an OpenAI-compatible endpoint, so any tool that lets you set a base URL works: point it at the local address, pick the model id from the models page, and leave the API key empty. Latency stays predictable because nothing leaves the machine.