An AI multi-agent orchestration architecture that enforces mandatory institutional review and provides real-time task monitoring and intervention. Full operation requires a pre-installed OpenClaw platform and a large language model API key.
Project overview
The architecture mandates plan rejection before execution and provides real-time task intervention (pause, cancel, resume), which distinguishes it from standard optional human-in-the-loop approaches.
Project type
AI Agent · Workflow
Deployment
Refer to project documentation
License
MIT
Best for
Developers, AI engineers, and enterprise teams who require mandatory institutional review, real-time monitoring, and task intervention for multi-agent orchestration.
Enterprise teams requiring real-time observability, audit logs, and intervention capabilities for multi-agent results.
Key capabilities
A web dashboard providing a Kanban board for task status, agent health monitoring, and task intervention (pause/cancel/resume).
Supports adding, viewing, and removing remote or local skills for specific agents via UI or CLI.
Limitations and risks
Full operation requires the OpenClaw platform to be pre-installed.
Agent operations require an LLM API Key, indicating a paid service dependency.
The data boundary is not documented; local orchestration does not prove a local-only data boundary without remote inference.
Getting started
Requires OpenClaw installed and Python 3.10+, involving running install.sh and start.sh scripts.
Clone the repository, run ./install.sh, and run ./start.sh.
Alternatives and comparisons
An open-source super agent harness that researches, codes, and creates using sub-agents, memory, sandboxes, and extensible skills.
Provides a Kanban UI to plan work and runs multiple coding agents in isolated workspaces, allowing users to review diffs and ship code faster.
An AI development environment that runs CLI coding agents side-by-side in isolated parallel worktrees, tracked from one app.