Hooman - Own your AI agent stack — local, open source and extensible

Hey Product Hunt 👋 I built Hooman because I wanted an AI agent stack I could actually own. Most agent tools are either tied to a hosted platform, limited to one interface, or difficult to adapt to private and local infrastructure. Hooman is an MIT-licensed, local-first agent runtime that works across the terminal, VS Code, ACP-compatible editors, and event-driven daemon workflows—all sharing the same configuration, sessions, skills, MCP tools, and approval model. You can use hosted providers, bring your own keys, point it at private inference endpoints, or run locally with llama.cpp and MLX. There’s no account requirement and no telemetry; your keys, sessions, and configuration stay on your machine. It also includes Agent, Plan, Ask, and Design modes, reusable skills, subagents, human-in-the-loop approvals, MCP integrations, background tasks, and design exports to formats such as PDF and PowerPoint. I’d especially love feedback on: • What would make you switch from your current coding agent? • Which model or provider integrations should I prioritize next? • Where does the setup or documentation still feel unclear? Thanks for checking it out. Hooman is open source, so issues, stars, feedback, and contributions are all genuinely appreciated 🙌