CodePilot
Getting Started

Quick Start

Get CodePilot running and connect your first repository in five minutes.

This guide gets you from zero to chatting with a codebase in the shortest path possible. For detailed setup, see Installation.

Five-Minute Quickstart

Clone and Install

git clone https://github.com/lovesinghal31/codepilot.git
cd codepilot
pnpm install

Start Services

Start the database and cache infrastructure:

docker compose up -d

Start Ollama and pull the embedding model:

ollama pull nomic-embed-text
ollama pull qwen2.5-coder:3b

Configure Environment

Copy the environment files and set minimum required values:

cp .env.example .env
cp .env.api.example .env.api
cp .env.worker.example .env.worker
cp .env.web.example .env.web

At minimum, ensure DATABASE_URL and REDIS_HOST are set correctly in .env.api and .env.worker.

Run Migrations and Start

pnpm --filter @repo/db run db:generate
pnpm --filter @repo/db run db:migrate
pnpm dev

The dashboard is now available at http://localhost:3000.

Connect a Repository

  1. Log in via GitHub OAuth on the dashboard
  2. Install the CodePilot GitHub App on your account
  3. Select a repository to connect
  4. CodePilot will automatically begin ingesting the repository

The worker will clone the repository, parse files, generate embeddings, and store vectors. You can monitor progress in the dashboard.

Chat with Your Code

Once ingestion is complete, navigate to the repository page and start asking questions:

  • "What does the main API handler do?"
  • "How is authentication implemented?"
  • "Show me the database schema"
  • "What design patterns are used in this codebase?"

CodePilot retrieves relevant code chunks via semantic search and generates context-aware answers using the local LLM.

What Happens Under the Hood

When you connect a repository, CodePilot performs the following pipeline:

  1. Clone — The worker clones the repo using a GitHub installation token via simple-git
  2. Scan — Files are scanned and filtered by language (TypeScript, JavaScript, Markdown, JSON)
  3. Parse — ts-morph performs AST-aware analysis to identify functions, classes, and components
  4. Chunk — Code is split into meaningful chunks (function bodies, class definitions, component trees)
  5. Embed — Each chunk is embedded using nomic-embed-text (768-dimensional vectors)
  6. Store — Vectors are stored in PostgreSQL via pgvector with HNSW indexing for fast retrieval

When you ask a question:

  1. Your query is embedded using the same model
  2. pgvector performs cosine similarity search to find relevant chunks
  3. The top-matching chunks are fed to qwen2.5-coder:3b as context
  4. The LLM generates a response grounded in your actual code

Next Steps

Project Structure

Understand how the monorepo is organized.

How It Works

Deep dive into the architecture.

Requirements

System requirements and prerequisites for running CodePilot locally or in production.

Project Structure

Understand the CodePilot monorepo layout — apps, packages, and how they work together.

On this page

Five-Minute QuickstartClone and InstallStart ServicesConfigure EnvironmentRun Migrations and StartConnect a RepositoryChat with Your CodeWhat Happens Under the HoodNext Steps