HKUDS/OpenOPC

What it solves

OpenOPC allows users to build and operate a "personal AI-native company" to handle complex, real-world tasks across various industries. It solves the problem of coordinating multiple AI agents by automating the recruitment of role-specific employees, managing their collaboration through a structured organizational hierarchy, and ensuring the system learns from its outcomes to improve future performance.

How it works

OpenOPC operates through three primary mechanisms:

1. Self-Built: The system drafts an organizational chart based on the goal, defines necessary roles and reporting lines, and recruits agents from a talent pool or reuses experienced employees who carry context from previous projects.
2. Self-Run: Work is managed via a state machine and a Kanban-style workflow. A manager agent decomposes goals into a dependency graph (DAG) of work items, assigning them to roles. It handles blockers by either resolving them within the team or escalating them to a human owner.
3. Self-Grown: The system improves by attributing outcomes to specific roles based on user feedback. It distills execution traces into high-signal lessons stored in private experience profiles and promotes recurring lessons into shared playbooks for new hires.

Who it’s for

It is designed for individuals or teams needing end-to-end execution in verticals such as AI research, software development, financial investment, sales growth, content production, accounting, e-commerce, and education.

Highlights

• Automated Org Charting: Dynamically creates roles and reporting structures based on the specific task.
• Multi-Mode Execution: Supports both a single-agent "Task Mode" and a multi-agent "Company Mode."
• Visual Management: Includes an "Office UI" with an animated office map, Kanban boards, and organization editors to monitor agent activity in real time.
• Organizational Memory: Builds a library of playbooks and experience profiles so the AI company becomes smarter over time.
• Extensible Agent Integration: Compatible with various execution agents including OpenOPC Native, Codex, Claude Code, and Cursor.