Tabby CLI Cheatsheet - Self-hosted AI Coding Assistant Full Reference

Tabby targets one specific job: code completion with model weights, indexing, and the API all on your own machine, so an IDE plugin only has to point its completion endpoint at localhost, with no external database and no cloud account. This table groups installation and startup, models and indexing, and configuration and API, covering how tabby serve launches the service, how download fetches models, and when scheduler should trigger repository indexing. The judgment is straightforward: whether completion feels good depends on model choice and index quality rather than on the deployment itself, so get one baseline completion model running and confirm the latency is acceptable before investing in answer-engine context. Reach for it when an internal network cannot install hosted services, when a team should share one private completion backend, or when the answer engine needs context from the current repository. Against a general-purpose gateway such as LocalAI, Tabby is tuned for editor completion and gives up broad chat capability in exchange.

AI CLI·20 commands·Last updated 2026-09-10

Install and Start 6

./tabby serve --model StarCoder-1B

Start with a model (CPU)

./tabby serve --model StarCoder-1B --device cuda

Start with NVIDIA GPU

./tabby serve --model StarCoder-1B --device metal

Start with Apple Silicon

./tabby serve --model StarCoder-1B --device cpu

Force CPU inference

./tabby serve --model StarCoder-1B --port 8080

Set the server port

./tabby --help

List all commands

Models and Indexing 6

./tabby download --model StarCoder-1B

Pre-download a model

./tabby scheduler --now

Trigger repo indexing now

curl http://localhost:8080/health

Check server health

docker run -p 8080:8080 tabbyml/tabby serve --model StarCoder-1B

Run via Docker

./tabby serve --chat-model Qwen2-1.5B-Instruct

Set a chat model

./tabby serve --model StarCoder-1B --chat-model Qwen2-1.5B-Instruct --device cuda

Completion plus chat together

Config and API 8

cat ~/.tabby/config.toml

View the config file

curl -X POST http://localhost:8080/v1/completions -d '{"language":"python","segments":{"prefix":"def "}}'

Call code completion

curl -X POST http://localhost:8080/v1/chat/completions -d '{"messages":[{"role":"user","content":"hi"}]}'

Call chat completion

./tabby serve --model TabbyML/DeepseekCoder-6.7B --device cuda

Use a larger completion model

./tabby serve --help

serve flags

./tabby download --help

download flags

./tabby scheduler --help

scheduler flags

curl http://localhost:8080/v1/models

List available models

Tips

Tabby exposes an OpenAI-compatible endpoint; point the IDE plugin at port 8080 to go private.
--device picks acceleration: cuda for NVIDIA, metal for Apple, cpu as fallback.
scheduler --now triggers code repo indexing so the answer engine gets project context.

FAQ

How is Tabby different from Continue?

Tabby is a self-hosted completion engine with its own server and a standard API; Continue is an IDE extension that can use many backends including Tabby, Ollama, or cloud.

Is a GPU required?

Not required. --device cpu works, just slower; cuda for NVIDIA and metal for Apple give a better experience.

Official References

Each command links to its official documentation below, so you can verify the latest usage and read deeper.

Tabby Docs

Maintained by LaoHand

Publicly updated on Sep 10, 2026, continuously proofread against official docs.

Contact Us

Wrong command or description? Send us corrections, business inquiries or product feedback by email.