FauxPilot

FauxPilot is an open-source, locally hosted AI coding assistant created as an alternative to GitHub Copilot. The project was introduced in 2022 by Brendan Dolan-Gavitt, a computer science professor at NYU Tandon, with the goal of providing AI-assisted code completion that could run on a user’s own infrastructure rather than relying entirely on a hosted service.

The project’s main focus is local AI-powered code generation and completion. FauxPilot can provide coding suggestions from a locally running server, allowing developers to use AI assistance while keeping their development environment under their own control. The project was specifically motivated by concerns around privacy, telemetry, and the need to send source code to third-party services.

FauxPilot originally used Salesforce CodeGen models running through NVIDIA Triton Inference Server with the FasterTransformer backend. The project later added support for CodeGen2 and a Python backend that broadened model compatibility.

A typical FauxPilot installation consists of a locally hosted server and a client or editor integration. The official repository provides Docker and Docker Compose configuration for running the inference server and proxy components. The project requires an NVIDIA GPU with sufficient VRAM for the selected model, making hardware an important consideration when deploying it locally.

FauxPilot exposes an API interface that can be used by compatible clients. The project documentation describes OpenAI API-compatible access, a REST API, and other interfaces for connecting development tools to the local inference server.

The project can also be used with development environments such as Visual Studio Code. Community documentation demonstrates using FauxPilot with the official Copilot Vim plugin as well as VS Code integrations, although some of these integrations are community-driven rather than maintained as official first-party editor extensions.

Another important capability is model customization. FauxPilot is not restricted to one permanently fixed model. Its documentation and community discussions describe converting compatible Hugging Face models and fine-tuning CodeGen-based models for use with the system. This makes the project particularly interesting for developers and researchers experimenting with specialized code models.

Privacy is one of FauxPilot’s defining characteristics. Because the inference system can run locally or on-premises, organizations can keep coding requests within their own infrastructure. This can be useful for teams that have policies restricting source-code transmission to external AI providers.

FauxPilot is also relevant as a research platform. Its creator described the project as a way to experiment with code-generation models and eventually investigate models that could produce more secure code. Researchers and developers can therefore use the project as a foundation for testing locally hosted code-generation systems.

The project is released under the MIT License, making it open source and permitting developers to inspect, modify, and redistribute the software subject to the license terms.

FauxPilot is notably more infrastructure-oriented than modern hosted coding assistants. Running it requires Docker, an NVIDIA GPU, model files, and appropriate configuration. As a result, it is better suited to technically capable developers, researchers, and organizations that specifically need local deployment than to users looking for a simple plug-and-play coding assistant.

The project has also seen relatively limited recent development compared with newer local coding assistants. The repository’s latest highlighted activity includes changes from 2024, while open issues include requests concerning newer GPU support, documentation, and additional integrations. This suggests that FauxPilot should be viewed primarily as an open-source/local coding project rather than a rapidly evolving commercial coding platform.

Pricing

FauxPilot is free and open source. There is no official subscription plan or per-user SaaS pricing for the project. Users are responsible for providing the infrastructure required to run the system, including compatible hardware and model resources.

Pricing Model

Free / Open Source

Pricing Note

FauxPilot itself does not charge a subscription fee. The main costs come from the infrastructure required to operate it locally, particularly an NVIDIA GPU with enough VRAM for the selected model, along with electricity, storage, and other hosting costs where applicable.

Review

FauxPilot is an interesting project for developers who want to experiment with self-hosted AI code completion. Its strongest advantage is that it provides a genuine open-source alternative to cloud-based coding assistants, allowing the inference infrastructure to remain under the user’s control.

Its support for configurable models is another benefit. Developers can work with CodeGen-family models and compatible Hugging Face models, giving technically experienced users more flexibility than a conventional closed coding assistant.

The main limitation is its technical requirements. FauxPilot is not a lightweight browser-based service. Docker, NVIDIA GPU hardware, model configuration, and server setup are required, and the quality of suggestions depends substantially on the selected model and available hardware.

The project’s relatively limited recent development is also worth considering. Newer local coding assistants generally offer more polished interfaces, broader editor support, and more modern AI capabilities. FauxPilot remains valuable as an open-source and research-oriented project, particularly for developers interested in local deployment and experimentation.