Skip to Content
Getting StartedInstall and run

Install and run Skippy

Skippy consists of a backend-neutral CLI and a verified native runtime selected for the host, architecture, accelerator, and Skippy ABI. Install both parts before serving a model.

Download a release when you want to run a published version. Build from source when developing Skippy or testing the current checkout.

Download a release

Download the archive for your platform from the latest Skippy release , or use the GitHub CLI:

gh release download --repo general-inference/skippy

Verify and extract the archive, then place the skippy executable and its adjacent native-runtimes directory together. Confirm the installation before serving a model:

skippy --version skippy doctor

Build from source

Skippy uses Just for repository builds. From the repository root:

just build ./target/debug/skippy --version ./target/debug/skippy doctor

just build produces the CLI and a verified adjacent native runtime. The CLI automatically discovers target/debug/native-runtimes from that build.

A source checkout may require a newer native ABI than the latest release. Use --runtime-bundle /path/to/bundle when you need to select a compatible local bundle explicitly.

How runtime discovery works

The CLI first checks verified runtime bundles beside the executable, then its configured bundle roots and platform cache. It admits a runtime only when its host, architecture, backend, release metadata, native ABI, and recorded file hashes are compatible with the CLI and serving request.

Run skippy doctor when selection fails. It reports the current machine, installed runtimes, supported backends, and the reason a candidate was rejected.

Serve a model

skippy models recommended skippy serve --model Qwen3-0.6B-Q4_K_M

Wait for the ready message or, in JSONL mode, the ready event before sending requests. The default public address is http://127.0.0.1:9337.

Confirm the model

curl http://127.0.0.1:9337/v1/models

A successful response confirms that the server is ready and exposes the model identifier clients must send in inference requests.

Command reference

TaskCommand
Download the latest releasegh release download --repo general-inference/skippy
Build the development productjust build
Inspect the installationskippy doctor
List installed runtimesskippy runtime list
Install a compatible runtimeskippy runtime install
Start servingskippy serve --model MODEL
Inspect served modelsGET /v1/models

Next: make an inference request.

Last updated on