FauxPilot

What it does

API integration

Clients can connect through an OpenAI-compatible API, though the full API is not implemented, or through a REST endpoint.checked 5 Oct 2026

Deployment

The server is started with Docker Compose after a setup script downloads and converts the selected model.checked 5 Oct 2026

Documentation status

The wiki says it is intended to collect usage, troubleshooting, and extension documentation and notes that its contents are limited.checked 5 Oct 2026

Editor integration

The documentation describes configuring the official VS Code Copilot plugin to use a locally hosted FauxPilot server.checked 5 Oct 2026

Hardware

The project requires an NVIDIA GPU with Compute Capability 6.0 or higher, and documented model VRAM needs range from 2GB to 32GB.checked 5 Oct 2026

Integration limits

The documentation says some Copilot plugin functionality may be subtly broken, returned probabilities are partly fake, and tokenizer differences can cause overlong requests.checked 5 Oct 2026

Intended users

The project describes itself as a locally hosted Copilot alternative and provides setup instructions for operating a server and connecting clients.checked 5 Oct 2026

License

The repository provides the software under the MIT License, free of charge and without restriction on use, copying, modification, distribution, or sale, subject to its conditions.checked 5 Oct 2026

Model choices

The setup documentation lists CodeGen models from 350M to 16B parameters, in mono-language Python-only and multi-language variants.checked 5 Oct 2026

Models

It uses Salesforce CodeGen models with NVIDIA Triton Inference Server and the FasterTransformer backend.checked 5 Oct 2026

Purpose

FauxPilot is an open-source, locally hosted alternative to the GitHub Copilot server.checked 5 Oct 2026

Support

The README points users to the wiki and discussion forum for questions and says there is no formal support or warranty.checked 5 Oct 2026