Installation

Install

hara is published as @nanhara/hara and requires Node.js 22.12 or newer. Older Node releases exit with an upgrade instruction instead of failing inside a dependency. The standalone installer and Hara Desktop include their own runtime.

npm i -g @nanhara/hara
hara --version

If your Node version is too old:

nvm install 22 && nvm use 22
npm i -g @nanhara/hara@latest

Connect a model

hara is not a model provider — it runs on a model you bring. This is the personal path. The easiest way is the interactive setup:

hara setup

It walks you through three choices:

1. Provider — anthropic · openai · qwen, or any OpenAI-compatible endpoint (GLM, DeepSeek, a local model…) via openai + a base URL.
2. API key — from that provider's dashboard.
3. Model — a sensible default is pre-filled; press Enter to accept.

Prefer one line (or scripting it)?

hara profile add personal --byok --provider anthropic --key sk-ant-... --model claude-opus-4-8
# any OpenAI-compatible endpoint (GLM / DeepSeek / local):
hara profile add personal --byok --provider openai --base-url <url> --key sk-... --model <model>

No key? The free Qwen tier signs in through your browser (no card — you just need a Qwen account):

hara login qwen

This opens a chat.qwen.ai link; sign in to your own Qwen account and approve (a browser sign-in, not a QR scan).

Setting hara up for a whole team? See the Organization Edition — your org runs a gateway and members join by URL, with no personal keys.

First run

Point hara at a project and talk to it in the TUI:

cd your-project
hara                 # interactive REPL
# or one-shot, headless:
hara -p "fix the null check in src/login.ts and run the tests"

hara doctor checks your provider, auth, model, roles, skills, and memory if anything looks off.