Skip to content
Hooman

Getting Started

Hooman is a hackable, local-first AI agent toolkit for building CLI, ACP, MCP, and channel-driven workflows. It’s built with TypeScript, the Strands Agents SDK, and Ink.

Everything — configuration, API keys, sessions, skills — lives in a single ~/.hooman directory shared across every surface. No account, no telemetry.

Pick your path

Section titled “Pick your path”
Terminal / CLI →

npx or bunx and you're chatting — no clone, no build.

 VS Code →

A native chat panel in the activity bar, from the Marketplace.

Requirements

Section titled “Requirements”
  • Node.js >= 24 or Bun >= 1.1
  • Credentials or a local runtime for whichever LLM provider you choose — see Models

By default Hooman runs Qwen3 1.7B locally via llama.cpp, with Qwen3.5 0.8B and Gemma 4 E2B preconfigured alongside it, so it works out of the box without any API keys — models download from the Hugging Face Hub on first use, with live progress (percent, size, speed, ETA) shown in every surface. Swap in a hosted provider whenever you’re ready.

What’s next

Section titled “What’s next”
  • CLI quickstart — install and run your first hooman chat session.
  • VS Code quickstart — install the extension and chat from the activity bar.
  • CLI reference — chat, exec, daemon, and /config in depth.
  • Configuration — ~/.hooman layout and repo-local overlays.
  • MCP — tool servers, OAuth, and event-driven automation.