Install guide · ~90 seconds
Instalar Liminal AI
Aplicativo desktop ou CLI com um comando. Windows, macOS e Linux. Licença FSL-1.1-MIT gratuita; traga sua própria chave API ou faça login para inferência gerenciada.
Desktop app
Download the native app
Download the native Liminal AI shell for your OS. It bundles the harness sidecar (liminald). you still need Node.js 20+ on PATH and an API key (or Vireon sign-in).
Windows
x64 · .zip
Unzip and run liminal_desktop.exe (keep liminald\ beside the exe).
liminal-desktop-windows-x64-v0.1.2.zip
macOS
Apple Silicon (arm64) · .zip
Unzip liminal_desktop.app → Applications (or run in place).
Unsigned alpha: right-click → Open if Gatekeeper blocks the first launch.
liminal-desktop-macos-arm64-v0.1.2.zip
Linux
x64 · .tar.gz
Extract and run ./liminal_desktop from the bundle folder.
liminal-desktop-linux-x64-v0.1.2.tar.gz
After download: first run
- Copy
liminald/repo/.env.exampletoliminald/repo/.envand setAGENT_API_KEY, or sign in with Vireon in the app. - Launch the desktop app from the extracted folder (do not move the binary without its data).
- Developer path: use the one-line CLI installer below for terminal or browser UIs: same harness, different shell.
All builds and checksums: GitHub Releases (v0.1.2-desktop)
Step 1
Or install via CLI
Prefer to audit first? View the install script source on GitHub: the hosted copy matches main.
Override install paths
Set these before running the installer:
LIMINAL_INSTALL_DIR. full path to the repo cloneLIMINAL_HOME. parent dir ($LIMINAL_HOME/liminal-ai)LIMINAL_SKIP_LAUNCH=1. skip auto-launch of the web UIAGENT_API_KEY. non-interactive CI install
Step 2
Choose a provider
The wizard asks for an OpenAI-compatible API key and base URL. Liminal works with any provider that speaks the protocol. We default to OpenRouter because it lets you swap models without rewriting code.
OpenRouter
Recommended. Hundreds of models behind one API. Pin DeepInfra for prompt-cache discounts.
- base
- https://openrouter.ai/api/v1
- model
- deepseek/deepseek-v4-pro
OpenAI
Direct. Set AGENT_MODEL to gpt-5 or whatever you have access to.
- base
- https://api.openai.com/v1
- model
- gpt-5
Anthropic
Via OpenAI-compat shim. Best paired with the fast model on a cheaper provider.
- base
- https://api.anthropic.com/v1
- model
- claude-opus-4-7
Local (LM Studio / Ollama)
Run entirely offline. Point AGENT_API_BASE_URL at your local server.
- base
- http://localhost:1234/v1
- model
- your-local-model
Step 3
Verify
liminal doctor checks Node 22+, npm 10+, .env API key, built artifacts, and port availability. Run it whenever something feels off.
liminal doctorOptional
Pro, Team, or Enterprise
Community Edition never needs a Vireon account. Pro and Team add cloud memory sync, license keys, and billing through a free Vireon account. install first, upgrade when you are ready.
Pro path: managed inference (no API key)
- Subscribe on the pricing page, then run
liminal login. - Skip OpenRouter in the install wizard if you only use Pro: the harness routes through Vireon with ~$5/mo included credits (pass-through).
- Top up anytime on Account → Managed inference.
Community path: bring your own key
Continue with Step 2 below: add AGENT_API_KEY from OpenRouter, OpenAI, or a local endpoint. No Vireon account required.
Step 4
Launch
Desktop (recommended): run liminal_desktop.exe (Windows) or liminal_desktop from the extracted bundle. First launch walks through persona bootstrap: voice, behavior, and UI theme.
liminal web --bootstrap --open
# developer path; desktop app is preferredDeveloper CLI
liminal commands
liminal setup
Interactive first-run wizard (writes .env, install, build)
liminal doctor
Verify Node, build artifacts, API key, port
liminal web --bootstrap --open
Web UI + open browser
liminal tui --bootstrap
Terminal UI
liminal update
git pull + npm install + build
liminal path
Print install directory
Stuck?
Run liminal doctor first. Most install issues fall into a handful of buckets: stale Node, missing build artifacts, port conflict, or a typo in the provider URL. The doctor explains each one.
Install FAQ
Installing Liminal AI
How do I download the Liminal Desktop app?
On the Install page, use the Desktop app section for your OS (Windows zip, macOS Apple Silicon zip, Linux tar.gz). Each build is on GitHub Releases (tag v0.1.2-desktop). You still need Node.js 20+ on PATH and an API key in liminald/repo/.env, or sign in with Vireon inside the app.
How do I install Liminal AI?
Download the desktop app for your OS in the section above (Windows zip, macOS arm64, Linux tar.gz from GitHub Releases). For the CLI path: curl -fsSL https://www.vireondynamics.com/install/install.sh | bash on macOS/Linux/WSL or irm https://www.vireondynamics.com/install/install.ps1 | iex on Windows: clones Liminal, runs setup, and builds the harness. You need Node.js 22+ and an OpenAI-compatible API key.
What are the system requirements for Liminal AI?
Node.js 22+ and npm 10+ on Windows, macOS, or Linux, plus an API key from any OpenAI-compatible provider (OpenRouter, OpenAI, Anthropic, or a local model via LM Studio or Ollama). No GPU required: the model runs at your provider, or locally if you point Liminal at a local endpoint.
Do I need an account or credit card to install Liminal?
No. Community Edition installs with no Vireon account and no credit card. You only need an API key for your chosen model provider. Optional Pro and Team plans use a free Vireon account, but you can install and use the agent fully without one.
Which API provider should I use with Liminal?
We default to OpenRouter because one key gives you hundreds of models and prompt-cache discounts on DeepInfra-backed routes. You can also use OpenAI, Anthropic, or a fully local model through LM Studio or Ollama by pointing AGENT_API_BASE_URL at your endpoint.
Liminal won't start: how do I fix it?
Run `liminal doctor`. It checks Node 22+, npm 10+, your .env API key, built artifacts, and port availability, and explains each failure. Most install issues are a stale Node version, missing build, a port conflict, or a typo in the provider URL.