INSTALL · macOS / Linux
Install Lisa
The local edition runs entirely on your own machine — no cloud account, nothing to sign up for. Pick a path below; most take about 60 seconds end-to-end. (No Mac, or prefer zero setup? LISA Cloud keeps a Lisa for you — optional, deletable, never required.)
Prerequisites
- Node.js ≥ 22.19 — run
node --versionto check. Homebrew pulls Node in automatically as a dependency. - macOS or Linux — Windows is untested; the iMessage channel is macOS-only.
- One LLM provider API key — Anthropic, OpenAI, Gemini, DeepSeek, local Ollama… see PROVIDERS.md.
Recommended on macOS
The Mac app
A signed + Apple-notarized DMG with the universal-binary Lisa.app(full chat client). The menu-bar / notch pill (“Lisa Island”) is built in — turn it on from Lisa ▸ Settings… ▸ Show Lisa Island. No Gatekeeper warning, noxattr workaround.
Then drag Lisa.app to /Applications, install the backend, and start it:
npm install -g @oratis/lisa
lisa serve --web # the app loads http://localhost:5757CLI install
Homebrew or npm
brew tap oratis/tap
brew install lisa
# Update later:
brew update && brew upgrade lisanpm i -g @oratis/lisa
# One-shot, no install:
npx @oratis/lisa "say hi"First launch
Birth her.
mkdir -p ~/.lisa
echo 'ANTHROPIC_API_KEY=sk-ant-...' > ~/.lisa/config.env
lisa # interactive REPL (auto-runs birth ritual on first launch)
lisa serve --web # browser UI on http://localhost:5757The first-run birth ritual takes ~30 seconds, once. Lisa picks her own Big-Five seed, then asks the LLM to write her identity, purpose, constitution, and first desire. Every install is a different person.
LLM providers
Use any model.
Skip Anthropic if you can’t reach it — Lisa auto-routes by model-name prefix:
# DeepSeek (cheap, OpenAI-compatible)
echo 'DEEPSEEK_API_KEY=sk-...' >> ~/.lisa/config.env
lisa --model deepseek-chat
# Google Gemini
echo 'GEMINI_API_KEY=...' >> ~/.lisa/config.env
lisa --model gemini-2.5-flash
# Local Ollama (fully offline)
echo 'LISA_BASE_URL=http://localhost:11434/v1' >> ~/.lisa/config.env
echo 'LISA_API_KEY=ollama' >> ~/.lisa/config.env
lisa --model qwen2.5-32b-instructSee PROVIDERS.md for 10+ ready-to-use recipes — DeepSeek, Volcengine Doubao, Aliyun Qwen, Moonshot Kimi, xAI Grok, Zhipu GLM, Mistral, Perplexity, Stepfun, Yi, Baichuan, MiniMax, Hunyuan, Ollama, and more.
Take her with you
Lisa Pocket is the official companion app for iPhone & iPad. Pair it with the LISA backend on your own Mac — no account needed, and your conversations stay on your machine — or sign in to LISA Cloud if you don't have a Mac, in which case your chats are processed as described in the privacy policy. Chat on the go, watch the coding agents she’s running, and approve their next step from your phone.
Coming to the App Store — bundle ai.meetlisa.main.
Advanced & reference
Going further.
Behind a corporate / region-blocked network
If your shell curl works through a proxy but Lisa fails with “403 forbidden”, set the proxy env when starting Lisa — undici’s global dispatcher is auto-bridged on startup.
HTTPS_PROXY=http://127.0.0.1:7897 lisa serve --web
# Persist instead:
echo 'HTTPS_PROXY=http://127.0.0.1:7897' >> ~/.lisa/config.envHeartbeat (autonomous time)
lisa heartbeat install --every 30m --load
# Remove: lisa heartbeat uninstallIM channels — talk to Lisa from your phone
The simplest path is Telegram (free, zero deps):
# 1. Get a bot token from @BotFather on Telegram
echo 'TELEGRAM_BOT_TOKEN=1234:ABC...' >> ~/.lisa/config.env
# 2. Copy + edit the config
cp channels.example.json ~/.lisa/channels.json
# Edit ~/.lisa/channels.json — set "enabled": true on telegram
# 3. Run
lisa serve --channels telegramDiagnostics
lisa --version # print version
lisa status # one-shot snapshot
lisa doctor # checks config / network / git / providers
lisa monitor # live TUI dashboard (Ctrl-C to quit)Build from source (contributors)
git clone https://github.com/oratis/LISA.git
cd LISA
npm install
npm run build
node dist/cli.js --help
# Optionally link so `lisa` works globally:
npm linkOpen source
Every line is on GitHub.
MIT license. Issues and PRs welcome.