Run a local coding agent with LM Studio + OpenPilot

Описание и применение

LM Studio упрощает настройку локальных моделей для OpenPilot, позволяя быстро подключать модели Qwen 3, увеличивать контекст и запускать сервер, после чего OpenPilot может читать, редактировать файлы и выполнять команды с подтверждением пользователя.

Задача

Need a local coding agent that can edit files, run shell commands, and ask for approval before sensitive actions, without relying on external terminals or cloud models.

Результат

LM Studio streamlines local model usage with OpenPilot, reducing setup steps, providing memory‑aware context management, and enabling quick, safe coding assistance.

Рабочий процесс

• 1. In LM Studio’s model search, download a Qwen 3 model (e.g., Qwen 3 8B or 27B).
• 2. Load the model and increase the context length above the default to accommodate file contents and tool results.
• 3. Start the local server via the Developer tab or run lms server start; note the base URL (default http://localhost:1234/v1).
• 4. Install OpenPilot (Windows: .exe, macOS: .dmg, Linux: clone, npm install, npm start).
• 5. On first launch, connect the model in OpenPilot: set Name to the LM Studio model identifier, Base URL to http://localhost:1234/v1, and API key to a placeholder or LM Studio token if authentication is enabled.
• 6. Pick a folder (preferably git‑tracked) and issue a concrete task, e.g., “Look at package.json and add a Scripts section to README.md explaining each one.”
• 7. Review the file reads, edits, and shell commands presented as cards, approving or denying any sensitive actions.

Доказательства

OpenPilot is a small open‑source desktop app I built because I wanted something Cursor‑style that would run on whatever model I picked, including local ones.

Источник

LM Studio does most of the work.

Источник

When you load the model, raise the context length above the default.

Источник

Open the Developer tab and switch on "Start server", or from a terminal: lms server start

Источник