Back to docs
Recipe

Recipe: In-IDE Code Copilot

Build a privacy-first AI coding assistant that runs inside your editor. No telemetry, no cloud dependency — your code never leaves your machine.

Overview

Local-First

Runs entirely on-device. Your source code is never transmitted to any external service.

Low Latency

Sub-200ms completions with streaming token generation for a fluid typing experience.

Multi-Language

First-class support for TypeScript, Python, Rust, Go, and C++ with context-aware suggestions.

Architecture

VS Code / JetBrains

Extension Host

Local Inference Server

llama.cpp / Ollama

FIM Engine

Fill-in-the-Middle

The extension captures cursor context and surrounding code, sends a FIM-formatted prompt to a local LLM server, and streams the completion back into the editor.

Quick Start

1

Install Ollama

brew install ollama # macOS / Linux
2

Pull a code model

ollama pull deepseek-coder-v2:16b
3

Install the extension

code --install-extension meridian.copilot

Configuration

Add to your VS Code settings.json:

{
  "meridian.copilot.provider": "ollama",
  "meridian.copilot.endpoint": "http://localhost:11434",
  "meridian.copilot.model": "deepseek-coder-v2:16b",
  "meridian.copilot.maxTokens": 256,
  "meridian.copilot.temperature": 0.2,
  "meridian.copilot.triggerMode": "auto"
}

Privacy Guarantees

  • No network egress beyond localhost — all inference runs on your hardware.
  • Zero telemetry collection — no usage data, no crash reports, no analytics.
  • Open-source extension — every line of the client is auditable on GitHub.
  • No account required — download, configure, and start coding immediately.

Ready to build?

Clone the starter kit and have a working copilot in under 10 minutes.

Get Started