sandydasari/openacme

OpenAcme is a platform for running a group of AI agents on your own laptop, organized like a small company. Instead of chatting with a single AI assistant, you set up multiple named agents, each with its own role, persona, memory, and choice of AI model. You decide how they relate to each other: a flat group of specialists you talk to directly, a manager agent that breaks down your requests and assigns pieces to others, or a layered structure with team leads and individual contributors. The core idea is that you hand the group a goal, and agents divide the work, track dependencies, and run in parallel where they can. Decisions they cannot make on their own surface as tasks waiting for your input. You answer those and the rest continues without you. For simpler tasks you can skip the hierarchy entirely and send a request straight to one specific agent. Each agent is stored as a folder on disk with a plain text file describing its role and persona, a workspace, and its own memory that persists across sessions. You can add or remove agents at any time, switch which AI provider or model one uses, or reshape its persona. A shared task board keeps every agent coordinated, and a SQLite database stores the full history of tasks, comments, and events so you can query or back it up directly. The tool runs as a background service on macOS and Linux. It registers with the system so it starts automatically at login. A web interface shows which agents are active, who is waiting for you, what tasks are on the board, and chat histories with each agent. You can also use a terminal interface if you prefer. You connect it to AI providers you already have accounts with, including Anthropic, OpenAI, Google, and others. Nothing leaves your machine except the prompts you send to whichever model provider you chose. The project is written in TypeScript and distributed as an npm package.