FauxPilot
A self-hosted server that aims to stand in for GitHub Copilot, running code models on your own GPU.
- Stars
- 14.7k
- Forks
- 637
- Licence
- MIT
- Language
- Python
- Last update
- 2.6 years ago
- Type
- Dev tool
No commits in over a year — check for forks or alternatives before adopting.
What it is
FauxPilot is an attempt to build a locally hosted alternative to GitHub Copilot. It runs SalesForce CodeGen models on NVIDIA's Triton Inference Server with the FasterTransformer backend, so code suggestions are generated on your own hardware instead of a cloud service. People use it when they want Copilot-style completions without sending code to a third party.
- Run a local code completion server on a machine with an NVIDIA GPU
- Point a client that speaks the OpenAI API or REST API at the local server
- Split a larger model across multiple GPUs, as the README describes for two RTX 3080 cards
Great for
- Developers with an NVIDIA GPU who want to keep their code on their own machines
- Teams experimenting with self-hosted code completion
- Users who already run Docker and are comfortable with server setup
- People who want to connect editors or tools through an OpenAI-style or REST API
Not ideal for
- People without an NVIDIA GPU, since the README lists one as a requirement
- Users who want a polished, supported product; the README states there is no formal support or warranty
- Anyone needing a guaranteed-current model lineup, since the last push shown is April 2024
How to get it
- 1
Clone and build from source
You'll need: Docker · docker compose >= 1.28 · NVIDIA GPU with Compute Capability >= 6.0 · nvidia-docker · curl and zstd
Clone the repository
git clone https://github.com/fauxpilot/fauxpilot - 2
Run it
- Confirm you have Docker, docker compose 1.28 or newer, and nvidia-docker with a supported NVIDIA GPU
- Run the setup script to choose and download a model
- Follow the server setup guide in documentation/server.md
- Configure a client using documentation/client.md
- 3
Get the most out of it
- Use the setup script to pick a model size that fits your total VRAM
- Split a model across multiple GPUs if you have more than one card
- Connect through the OpenAI API, Copilot plugin, or REST API options described in the client guide
- Check the wiki and discussion forum linked in the README for extra information
Downloads
Strengths
- Runs locally, so code does not have to leave your own infrastructure
- Uses established components: Triton Inference Server and FasterTransformer
- Open source under the MIT license with a large following (14,668 stars shown)
Trade-offs
- Requires an NVIDIA GPU with Compute Capability 6.0 or higher and enough VRAM for the chosen model
- Setup involves Docker, nvidia-docker, and model download and conversion steps
- The README gives no formal support, and the last push shown is from April 2024
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| fauxpilot (this) | 14.7k | Dev tool | advanced | 2.6 years ago | Developers with an NVIDIA GPU who want to keep their code on their own machines |
| tabby | 75.3k | App | beginner | 4 days ago | Developers who keep many SSH sessions open and want them saved and organized |
| ollama | 186k | App | beginner | 4 days ago | Developers who want a local model API for their own apps |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.