Features

One engine, two places to run it.

The same scan, diagnose, repair, review, and generate functions run in the CLI, the desktop app, and this hosted desk. Hosted plans add queueing, history, GitHub, and larger models. They do not replace Ollama on your computer.

Free local models

Ollama is the default. LM Studio, any OpenAI-compatible server, Groq, OpenRouter, and Hugging Face are optional. A GGUF file is real only after you download it.

Safe repairs

Hosted repair uses the engine's safe pass. It will not delete a module to silence an error, and it will not pretend a missing import exists.

Workers AI

Standard, Expert, and Extreme call different default models. You can override the ids with HOSTED_MODEL_STANDARD, HOSTED_MODEL_EXPERT, and HOSTED_MODEL_EXTREME.

Bring your own key

Every paid plan can store an OpenAI, Anthropic, OpenRouter, or Groq key. Keys are encrypted with AUTH_SECRET and are not written into run logs.

GitHub, when the plan allows

Expert connects a repository with a token you supply. Extreme can open a pull request containing the safe repair.

Missions and watchdogs

Extreme runs diagnose, safe repair, review, and generate as one mission. Expert schedules a weekly watchdog. Extreme can schedule a daily one.