How I Built a Fully Local AI Coding Agent with Qwen 3.8, LM Studio, and Pi (No Paid APIs)

Описание и применение

Автор создал полностью локальный AI‑агент, использующий модель Qwen 3.8 27B, LM Studio и Pi Coding Agent, чтобы писать, запускать и исправлять код без обращения к облачным API.

Задача

Build a fully local AI coding agent that can read files, write code, execute commands, and fix errors without sending any requests to external APIs.

Результат

The author successfully built a React + TypeScript Mission Control dashboard entirely locally, with the agent generating code, executing commands, testing, and fixing errors autonomously.

Рабочий процесс

• Install LM Studio and download the 4‑bit quantized Qwen 3.8 27B model
• Load the model in LM Studio and configure a 200,000‑token context length
• Enable LM Studio’s local OpenAI‑compatible API server (default port 1234)
• Install Pi Coding Agent from https://pi.dev/
• Install the pi‑lmstudio extension to discover LM Studio models
• Restart Pi and select the Qwen 3.8 27B model via /model
• Use Pi to issue coding prompts; Pi sends requests to the local LM Studio API, receives tool calls, and executes them locally

Доказательства

Using Qwen 3.8 27B, LM Studio, and Pi Coding Agent, I built a fully local AI coding workflow and challenged it to create an entire web application from scratch.

Источник

The 4‑bit model I downloaded was approximately 16GB.

Источник

LM Studio can expose your local model through an OpenAI‑compatible API.

Источник