Tabby — Self-Hosted Code Completion

Introduction

GitHub Copilot costs $19/user/month and sends your proprietary code to Microsoft's servers. For a team of 10, that's $190/month — and every line of your codebase is processed externally.

Tabby runs on your VPS, uses open models, and keeps every line of code private. The ROI case practically makes itself.

Tabby vs Continue.dev

Deploying Tabby

mkdir -p ~/ai-stack/tabby && cd ~/ai-stack/tabby

version: "3.8"

services:
  tabby:
    image: tabbyml/tabby:latest
    container_name: tabby
    restart: unless-stopped
    command: serve --model StarCoder-1B --device cpu
    ports:
      - "8080:8080"
    volumes:
      - tabby-data:/data
    environment:
      - TABBY_DISABLE_USAGE_COLLECTION=1

volumes:
  tabby-data:

docker compose up -d

Tabby dashboard available at http://your-server-ip:8080.

Model Selection for Code Completion

Remember to account for RAM used by Ollama (Part 1). On an 8GB VPS, use StarCoder-1B or DeepSeek-Coder-1.3B alongside Ollama.

IDE Integration

VS Code

1. Install the Tabby extension from the VS Code marketplace
2. Open Settings → search "Tabby"
3. Set Server URL to http://your-server-ip:8080
4. Set your authentication token (from Tabby dashboard)
5. Start typing — completions appear inline automatically

JetBrains IDEs

1. Go to Settings → Plugins → Marketplace
2. Search and install "Tabby"
3. Configure the server endpoint under Settings → Tools → Tabby
4. Works with IntelliJ, PyCharm, WebStorm, and all JetBrains IDEs

Repository Context

Tabby can index your codebase for context-aware completions that understand your project's patterns, naming conventions, and architecture:

1. Navigate to the Tabby dashboard → Repositories
2. Add your Git repositories (supports GitHub, GitLab, or local paths)
3. Configure indexing schedules (hourly/daily)
4. Select which repositories and branches to index

This dramatically improves suggestion quality — completions reference your actual code patterns rather than generic suggestions.

Team Configuration

Set up multi-user access:

• Create user accounts from the Tabby dashboard
• Generate per-user authentication tokens
• View usage analytics per developer
• Configure different models for different teams (Python team might prefer CodeLlama, TypeScript team might prefer StarCoder)

Performance Optimization

Completions should feel instant — aim for under 500ms latency:

Continue.dev Alternative

If you prefer a simpler setup that connects directly to Ollama (Part 1):

1. Install the Continue extension in VS Code
2. Open ~/.continue/config.json
3. Configure your Ollama endpoint:

{
  "models": [{
    "title": "Ollama - Mistral",
    "provider": "ollama",
    "model": "mistral",
    "apiBase": "http://your-server-ip:11434"
  }],
  "tabAutocompleteModel": {
    "title": "DeepSeek Coder",
    "provider": "ollama",
    "model": "deepseek-coder:6.7b",
    "apiBase": "http://your-server-ip:11434"
  }
}