FauxPilot
Self-hosted Copilot-style completion server on your own GPU — dormant since 2023
Overview
Open-source (MIT), self-hosted completion server with an OpenAI-style API that VS Code's Copilot extension can be pointed at, serving Salesforce CodeGen models on NVIDIA Triton. Code never leaves your hardware, but its main branch has been dormant since May 2023 and the base models are dated.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 28 Sep 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
Use cases
What you can produce with FauxPilot
- Self-host a Copilot-style completion API on your own NVIDIA GPU
- Serve Salesforce CodeGen models, 350M to 16B parameters, through NVIDIA Triton
- Keep code entirely on your hardware, which is the reason it existed
- Study it as a reference for building a private completion server
- Do not build on it in 2026: the main branch has had no commits since May 2023 and there are no published releases
- Pick a maintained self-hosted assistant instead — the base models are superseded too
FauxPilot proved in 2022 that you could run a Copilot-style completion server on your own hardware — Salesforce CodeGen models on NVIDIA Triton behind an OpenAI-style API that VS Code's Copilot extension could be pointed at — which made it a reference point for teams that wouldn't send code to the cloud. That concept won; this implementation didn't. The main branch has had no commits since May 2023, no releases were ever published, NVIDIA has ended development of the FasterTransformer backend it relies on, and its CodeGen base models are generations behind current open code models. For real self-hosted coding assistance now, Tabby, Continue, or an Ollama-backed setup is more current, easier to run, and actually maintained. FauxPilot is worth knowing as the category's proof of concept — not the tool to standardise on in 2026.
What we found
Don't build on this in 2026. FauxPilot's GitHub repo is still up — MIT-licensed, zero published releases — but its main branch has had no commits since May 2023, and the last work, a February 2024 branch adding CodeGen2 support, was never merged. The README's support section opens with the word "lmao" before pointing to a wiki and a discussion forum.
Which is a shame, because the idea was ahead of its time. In 2022, FauxPilot, written mainly by NYU security researcher Brendan Dolan-Gavitt, proved you could stand up a server with an OpenAI-style API, point VS Code's Copilot extension at it through a debug setting, and get completions from Salesforce CodeGen models running on NVIDIA's Triton Inference Server with the FasterTransformer backend — without a single line of code leaving your network. For banks, defence contractors and NDA-bound teams, that was the whole pitch, and it worked.
The problems are the base models and the plumbing. CodeGen was respectable in 2022 and is nowhere near current open code models like Qwen2.5-Coder or DeepSeek-Coder. Triton plus FasterTransformer is a heavyweight, fiddly stack that demands a serious NVIDIA GPU and real Docker patience, and NVIDIA itself has stopped developing FasterTransformer in favour of TensorRT-LLM. For the same privacy guarantee today, Tabby gives you a maintained self-hosted assistant, Continue plus an Ollama backend takes twenty minutes to set up, and both serve models years ahead of anything FauxPilot ships.
Credit where due: it was an early demonstration of the self-hosted Copilot category. But a dormant repo serving 2022-era models is a reference, not a tool.
Its setup script offers CodeGen models from 350 million to 16 billion parameters; the largest, Python-only, needs 32GB of GPU memory in total, and any card must be an NVIDIA GPU of compute capability 6.0 or above.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
Notable facts
- FauxPilot was created within weeks of GitHub Copilot's general availability, demonstrating how quickly the open-source community reacts to commercial AI tool launches.
- The project was named as a deliberate pun — 'faux' (fake in French) + 'Pilot', reflecting its nature as an open-source alternative to GitHub's Copilot.
- FauxPilot's architecture works by intercepting Copilot API requests in the IDE and redirecting them to a local server — no IDE modification required.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including FauxPilot's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with FauxPilot unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to FauxPilot directly →
Spotted something out of date? Suggest an update →
Alternatives to FauxPilot
More in Code & Dev Tools
