This project provides a fully autonomous AI company capable of running 24/7 on a user's local machine. It addresses the challenge of automating complex business operations, from ideation and product development to deployment and marketing, by orchestrating 14 specialized AI agents. The primary benefit is enabling continuous, hands-off business operations driven by sophisticated AI decision-making.

How It Works

The system operates via a continuous loop managed by daemons (launchd on macOS, systemd --user on WSL). In each cycle, AI agents read shared consensus memory, form expert teams (3-5 agents), execute tasks using LLM CLIs (Claude Code or Codex CLI), and update the consensus memory. This approach leverages expert persona-based prompting, mimicking world-class professionals, and relies on a single shared state (consensus.md) for coordination, enabling autonomous task execution and decision-making.

Quick Start & Requirements

• Primary Install/Run: Clone the repository, cd into it. On macOS/WSL, use make start (foreground) or make install (daemon). On Windows (WSL), use .\scripts\windows\start-win.ps1.
• Prerequisites: macOS or Windows + WSL2 (Ubuntu); Claude Code or Codex CLI installed and authenticated; Node.js; make (macOS/WSL); jq (recommended for WSL); gh and wrangler (optional); sufficient LLM model quota.
• Resource Footprint: Involves ongoing costs due to LLM API calls per cycle. Setup time depends on LLM CLI installation and environment configuration.
• Links: Codex CLI: https://www.npmjs.com/package/@openai/codex.

Highlighted Details

• Features 14 AI agents modeled after renowned experts (e.g., Jeff Bezos for CEO, DHH for Full-Stack).
• Operates autonomously 24/7, managing product ideation, coding, deployment, and marketing.
• Employs defined workflows (e.g., New Product Evaluation, Feature Development) with specific agent collaboration chains.
• Allows user intervention by editing consensus.md or using pause/resume commands.
• Includes safety guardrails defined in CLAUDE.md to prevent destructive actions.

Maintenance & Community

The project provides contribution guidelines (fork, branch, PR) but does not list specific maintainers, sponsors, or community channels like Discord or Slack. Acknowledgments reference related projects.

Licensing & Compatibility

The project is licensed under the MIT license, which generally permits commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

This is an experimental project with stability not guaranteed. Each operational cycle consumes LLM model quota, incurring costs. The autonomous nature means agents act without explicit human approval, necessitating careful configuration of guardrails in CLAUDE.md. Platform-specific setup is required for macOS (launchd) and Windows (WSL2 with systemd --user).