DEV Community

mtdnot
mtdnot

Posted on

I built a visual org chart editor for OpenClaw multi-agent systems

Design your AI agent team visually, export OpenClaw config in one click

The problem: 5 agents, zero visibility

I was running a multi-agent setup on OpenClaw β€” a PM agent, a dev agent, a researcher, a critic, and myself as the human owner. Five agents, all talking to each other through different channels.

Within a day, I had no idea who was reporting to whom. The PM was supposed to manage the dev, but the critic was giving direct orders. The researcher was duplicating work the dev had already done. Config files were a mess of JSON with agent IDs pointing everywhere.

Sound familiar?

If you've tried running more than 2-3 agents on OpenClaw, you've probably hit the same wall: there's no way to see your organization. You're editing raw config and hoping the relationships make sense.

What I wanted

Something dead simple:

  • See all my agents and their relationships in one view
  • Drag and drop to reorganize the hierarchy
  • Click to define who reports to whom, who reviews whom
  • Export a working OpenClaw config β€” not some abstract diagram, but actual config.apply-ready JSON

Basically: an org chart editor that speaks OpenClaw.

So I built AgentFlow

AgentFlow Editor

AgentFlow is a browser-based visual editor for designing OpenClaw multi-agent organizations. No backend, no accounts β€” it runs entirely in your browser.

What it does

Design your org visually:

  • Add agents with one click
  • Drag to arrange your hierarchy
  • Connect nodes to define relationships

Three relationship types:

  • πŸ”΄ Authority (solid line) β€” command chain, who manages whom
  • πŸ”΅ Communication (dashed line) β€” information flow between peers
  • 🟑 Review (dotted line) β€” who reviews whose work

Configure each agent:

  • Name, role, personality (β†’ SOUL.md)
  • Model selection (Claude, GPT, Gemini)
  • Tool profile (minimal / coding / messaging / full)
  • Initial memory (β†’ MEMORY.md)

Export to OpenClaw:
One click generates:

  • config.json β€” ready for openclaw config apply
  • setup.sh β€” creates agent workspace directories with SOUL.md and MEMORY.md
  • Full export JSON with metadata

Org templates

Don't want to start from scratch? Pick a template:

  • πŸš€ Startup β€” CEO β†’ PM β†’ DevΓ—2
  • 🏒 Hierarchy β€” Director β†’ Managers β†’ Team members
  • 🀝 Flat β€” Lead + peers, everyone communicates equally

Try it

Live demo (no install): frexida.github.io/agentflow/editor/

Self-host with Docker:

git clone https://github.com/Frexida/agentflow.git
cd agentflow
docker compose up -d
Enter fullscreen mode Exit fullscreen mode

GitHub: github.com/Frexida/agentflow

What's next

This is early β€” I'm looking for feedback from people who actually run multi-agent setups on OpenClaw. What's missing? What would make this useful for your workflow?

If AgentFlow looks interesting, ⭐ the repo and drop an issue with your thoughts. Every bit of feedback shapes what gets built next.


AgentFlow is MIT licensed and 100% open source. Built with Astro, Drawflow, and dagre.

Top comments (0)