Wally runs language models, local or in the cloud, and points your coding tools at them through one local server.
wally login # sign in to Wally Cloud in your browser
wally serve # start the local server other tools talk to
wally run llama-3.1-8b # run a model once, or open a chat if you leave off the prompt
wally opencode # point OpenCode at that model and launch itwally login is optional if you only want an on-device model, but on-device
inference itself is not wired into this build yet. See Status.
Build from source (see below), or use the installers once a release exists:
curl -fsSL https://raw.githubusercontent.com/RunanywhereAI/wally/main/install.sh | sh # macOS/Linuxirm https://raw.githubusercontent.com/RunanywhereAI/wally/main/install.ps1 | iex # Windowsinstall.sh and install.ps1 fetch a released binary from GitHub Releases.
git clone https://github.com/RunanywhereAI/wally.git
cd wally
./scripts/build.sh
./build/wally versionscripts/build.sh builds the single wally binary for your machine's OS and
architecture. scripts/dev.sh does the same but points the binary at the dev
console instead of production, useful when testing against a non-production
backend. scripts/test.sh runs the test suite plus formatting and vet checks.
See docs/commands.md and docs/daemon.md
for what each piece does.
Grouped the way wally help groups them:
| Group | Commands |
|---|---|
| Run | wally run [model] [prompt], wally chat |
| Serve | wally serve |
| Models | wally models list | show | rm | pull |
| Coding agents | wally claude-code, wally claude-desktop, wally opencode, wally hermes, wally openclaw, wally deepseek |
| Manage | wally harness, wally harness set-default-model |
| Account | wally login, wally logout, wally whoami |
| About | wally version, wally info, wally uninstall |
Run wally help or wally <command> --help for the full picture; the table
above is the shape, not the whole reference. docs/commands.md covers each
group in more detail.
This is the Go rewrite of the wally CLI. The command surface, the local daemon, cloud login, and the coding-agent launchers are live. Two pieces are still deferred:
- On-device inference.
wally runandwally chatwork against Wally Cloud today. The on-device engine link is not in this build, so a local model returns a "not enabled in this build yet" error instead of running. - Model download.
wally models pullneeds a downloader wired in viaWALLY_PULL_SCRIPTorscripts/pull-model.sh. Neither ships yet, so a pull fails with a clear message pointing at that gap.
See docs/ for what each area actually does today, deferred pieces called out
where they apply.
- docs/commands.md: the full command surface, grouped and explained
- docs/daemon.md: the local server every client talks to
- docs/harnesses.md: what a coding-agent launcher does and how one is wired
- docs/login.md: the device-flow login and where credentials live