Anil-matcha/open-duet

What it solves

Open Duet helps you build a team of AI assistants that you can name, give jobs, and chat with over time. It solves the problem of managing multiple AI agents with different roles, keeping context between conversations, scheduling work, and keeping control over what they do.

How it works

You create agents, each with a name, instructions, model choice, workspace, memory, and approval settings. Agents can work independently or as a team: a lead agent can delegate tasks to specialists, and they can share rooms. Agents can keep notes and recall relevant context. You can set routines to run on a schedule or via triggers. You choose which engine (like Claude Code, Codex CLI, Gemini, etc.) handles the prompts. Open Duet runs on your machine or a server you control, and stores data locally.

Who it's for

Open Duet is for individuals or small teams who want a personal or self-hosted AI workspace. It suits developers, researchers, writers, or anyone who wants to orchestrate multiple AI agents for tasks like research, planning, writing, or development. It requires some technical setup (Node.js, uv, and an engine) but is designed to be run on your own hardware.

Highlights

• Create a roster of named agents with distinct roles and instructions.
• Agents can work as a team: lead agents delegate to specialists.
• Built-in memory: agents keep notes and recall relevant context.
• Schedule routines to run on a schedule, via webhook, or on demand.
• Approval controls: choose between Ask first, Accept edits, or Plan only.
• Runs locally on your computer or a server you control; no hosted model included.
• Connects to external engines like Claude Code, Codex CLI, Gemini, or OpenAI-compatible APIs.
• Track token usage and costs, attach files, and search conversations.
• Experimental: server setup for single owner, Docker computers optional.