create-kriya-app

Scaffold a new local-first agent app — built on kriya, the governed runtime that lets an AI agent safely drive a desktop app — in one command.

npm create kriya-app@latest my-app
# or
npx create-kriya-app my-app

Already have an app? You don't need to start from scratch — bolt kriya onto your existing handlers with wrapAction (see kriya-core). This scaffolder is for greenfield apps.

You get a working Tauri 2 + React + TypeScript desktop app with the agent host (kriya-core SDK + Rust agent loop) already wired in: typed actions, permission policy, human-approval queue, action budget, signed audit, persistent memory, and four swappable inference backends (deterministic, claude-cli, ollama, anthropic).

The starter is a tiny counter app: humans click the buttons; the agent calls the same actions to reach a target count. reset_counter is gated behind human approval to show the approval flow.

Next steps

cd my-app
npm install
npm run tauri dev

First run compiles the Rust agent host — a few minutes. After that, click Run agent and watch it operate your app through typed actions.

Pick a backend with AGENT_BACKEND (deterministic is the default and needs no setup):

AGENT_BACKEND=claude-cli npm run tauri dev
AGENT_BACKEND=ollama OLLAMA_MODEL=llama3 npm run tauri dev
AGENT_BACKEND=anthropic ANTHROPIC_API_KEY=... npm run tauri dev

See the project README and architecture.md once scaffolded.