FauxPilot
Open-source self-hosted alternative to GitHub Copilot
About FauxPilot
FauxPilot is a self-hosted, open-source code completion server that replicates GitHub Copilot functionality using Salesforce CodeGen models on NVIDIA Triton Inference Server. It lets developers run AI-powered code suggestions entirely on their own hardware, ensuring source code never leaves local infrastructure.
Pros & Cons
Pros
- Complete data privacy — code never leaves your infrastructure
- No subscription cost under MIT license
- Drop-in Copilot replacement compatible with existing editor plugins
- Model flexibility to choose size based on hardware or swap custom models
- Offline capable for air-gapped or compliance-restricted environments
- Research-grade foundation built by NYU professor
Cons
- Requires NVIDIA GPU — no CPU-only or AMD GPU support
- Complex setup with Docker, nvidia-docker, model conversion, and Triton configuration
- No formal support — community-driven with no commercial backing
- Lower quality suggestions compared to commercial GitHub Copilot
- Limited editor documentation beyond VS Code
Key Features
Self-Hosted Code Completion
Runs entirely on local hardware, providing AI code suggestions without sending code to external servers
OpenAI API Compatibility
Exposes an OpenAI-compatible API endpoint for easy integration with existing tools
Copilot Plugin Support
Works with existing GitHub Copilot editor extensions by redirecting the endpoint URL
REST API
Standard REST API for custom integrations and tooling beyond IDE plugins
Multiple Model Sizes
Supports CodeGen models from 350M to 16B parameters based on available GPU VRAM
Multi-GPU Splitting
Large models can be split across multiple GPUs for running bigger models
Docker-Based Deployment
Docker Compose stack for reproducible, containerized setup
Multi-Language Models
Offers both Python-specialized and multi-language model variants
Pricing
- Full code completion server
- All model sizes
- OpenAI-compatible API
- Docker deployment
- MIT license
Best For
Enterprise Code Privacy
Organizations with strict data governance can get AI coding assistance while keeping all data on-premises
Air-Gapped Development
Teams in classified or regulated environments deploy offline AI code completion
AI Code Security Research
Academics study vulnerabilities in AI-generated code and develop safer models
Cost-Conscious Teams
Developers with NVIDIA GPUs avoid recurring Copilot subscription fees by self-hosting
Custom Model Experimentation
ML engineers swap in fine-tuned models to benchmark different code generation approaches
Similar Tools
Visual Studio Code
Free, open-source code editor from Microsoft
Chat2DB
AI-powered SQL client that turns natural language into database queries
Chroma
The open-source AI-native vector database for search and retrieval
Zed
The fastest AI code editor — built in Rust for speed and collaboration
Activepieces
Open-source, AI-first business automation
Blackbox AI
AI coding assistant with 300+ models and autonomous agents