Tabby Download: A Self-Hosted AI Coding Assistant

Tabby lets a developer or a whole team run a Copilot-style assistant on their own hardware, so code never has to leave the company network. It is self-contained, with no database or cloud service to set up, and offers an OpenAPI interface for connecting it to existing tools.

What does Tabby do?

• Code completion: Suggestions as you type, using context from your repository.
• Chat: A side panel for questions, with an @ menu to add context.
• Answer Engine: A knowledge engine for your team’s internal engineering information.
• Team features: Team management, secured access, an admin UI, usage reports and GitHub/GitLab integration, including self-hosted instances.
• IDE extensions: VS Code, Vim and JetBrains IDEs such as IntelliJ.

Who it is for

It is aimed at teams and companies that want AI code completion without sending source code to a third party, and at individuals who want to run their own.

Starting a Tabby server

With Docker and an NVIDIA GPU:

docker run -it \
  --gpus all -p 8080:8080 -v $HOME/.tabby:/data \
  tabbyml/tabby \
  serve --model StarCoder-1B --device cuda --chat-model Qwen2-1.5B-Instruct

On an Apple Silicon Mac, use Homebrew:

brew install tabbyml/tabby/tabby
tabby serve --device metal --model StarCoder-1B --chat-model Qwen2-1.5B-Instruct

Guides for other setups are in the official docs. The server listens on port 8080 by default (change it with --port); then install the Tabby extension in your editor and connect it.

Hardware requirements

• GPU: Consumer-grade GPUs are supported; the Docker example uses NVIDIA CUDA.
• Apple Silicon: M1/M2 Macs work through Metal, but the project says this is likely only enough for individual use. For a team server it recommends Docker with CUDA or ROCm.

License

Tabby is open source under the Apache 2.0 license; code in the repository’s ee/ (enterprise) directory has its own separate license.