Edict reimagines AI multi-agent orchestration by implementing a governance structure inspired by ancient Chinese imperial bureaucracy ("三省六部"). It addresses the common shortcomings of current frameworks—lack of rigorous review, auditability, and real-time control—by introducing institutional checks and balances, a comprehensive dashboard, and detailed audit trails. This system is designed for technically savvy users seeking a more robust, reproducible, and manageable approach to complex AI agent workflows.

How It Works

Edict structures agent collaboration using a hierarchical system mirroring the "三省六部" model. Agents are assigned roles such as "太子" (sorting incoming requests), "中书省" (planning and task decomposition), "门下省" (review and veto), "尚书省" (dispatching tasks), and specialized "六部" (execution). The core innovation lies in the "门下省," which acts as a mandatory quality assurance gate, reviewing and potentially rejecting plans before execution. This institutionalized review, coupled with a real-time dashboard and complete audit logs, provides a level of control and transparency absent in many contemporary multi-agent systems.

Quick Start & Requirements

A full dashboard demo with mock data can be launched via Docker: docker run -p 7891:7891 cft0808/edict. For manual installation, prerequisites include Python 3.9+ and Node.js 18+ (for frontend build). The process involves cloning the repository, running install.sh, and then starting the data refresh loop (scripts/run_loop.sh) and the dashboard server (dashboard/server.py). Access the dashboard at http://localhost:7891. Detailed documentation is available in docs/getting-started.md and docs/task-dispatch-architecture.md.

Highlighted Details

• Institutional Review: The "门下省" provides a mandatory quality gate, capable of vetoing subpar plans.
• Real-time Dashboard ("军机处"): Offers a Kanban view, task status monitoring, agent health checks, and intervention capabilities (stop, cancel, resume).
• Audit Trails ("奏折阁"): Automatically archives completed tasks with a five-stage timeline for full traceability.
• Agent & Model Management: Each agent has independent workspaces, skills, and configurable LLMs, with hot-swapping supported via the dashboard.
• Extensible Skills: Remote skills can be added via UI, CLI, or API, integrating with an official Skills Hub.
• Data Input Sanitization: Automatically cleans input decrees by stripping file paths, metadata, and prefixes.

Maintenance & Community

The project is built by the "OpenClaw Community" and includes a CONTRIBUTING.md file, indicating community involvement. Specific community channels or prominent maintainer details are not explicitly listed in the README.

Licensing & Compatibility

Edict is released under the MIT License, making it permissive for commercial use and integration into closed-source projects.

Limitations & Caveats

While the core architecture is functional, the roadmap indicates that advanced governance features like manual approval ("御批模式") and agent performance scoring ("功过簿") are still under development, suggesting ongoing maturation. The frontend build requires Node.js 18+, and the system relies on the separate installation of OpenClaw if not using the Docker image.