Real-world multi-agent fleet feedback — 9 agents in production

[Jbennett111]

Running a 9-agent OpenClaw fleet in production (consulting firm + automotive client ops). Currently built a 2D pixel office (Star Office) to visualize agent state, but Claw3D is exactly what we need. Here is feedback from deploying this pattern daily:

What matters most

1. Multi-agent visibility is the killer feature
We have 9 specialized agents (CEO/router, ops manager, engineer, data analyst, etc). Seeing Agent A route work to Agent B, who hands off to Agent C — that is what makes a 3D office 10x better than logs. Lean into the "office as operations center" angle.

2. Agent state to animation mapping should be config, not code
We use states like idle/writing/executing/syncing/error. Being able to define custom state-to-animation mappings via config (not source code changes) would make adoption much easier.

3. Mobile-first matters
We demo from phones constantly. The 3D office needs to work well on iOS Safari with touch/swipe navigation — not just responsive, but designed for mobile.

4. Sound design
Subtle ambient office sounds + audio cues when agents change state makes the experience feel alive. Even simple notification sounds when an agent starts/finishes a task.

5. Auth integration (Entra/SSO)
Enterprise customers need auth. We are currently running oauth2-proxy + Entra OIDC in front of our apps. Built-in OIDC support or documented oauth2-proxy integration would help adoption.

6. External event webhooks
Let external systems push state into the 3D world. Examples: HubSpot deal closes = confetti in the office. CI build fails = alarm in the server room. New client onboarded = doorbell rings.

Our setup

• 9 OpenClaw agents with role isolation
• Azure VM + nginx + Lets Encrypt + Entra auth
• Star Office (2D pixel office) showing agent states
• Dashboard + Paperclip for fleet management
• HubSpot, Azure SQL, M365 integrations

Happy to be an early tester when code drops. This is the right vision.

[gsknnft]

This is extremely useful feedback. Thank you.

A lot of what you called out lines up with where Claw3D is heading now, and we’ve already started laying the groundwork for it in the current roadmap.

The big one is your point about multi-agent visibility being the killer feature. That is exactly the direction we want to lean into: Claw3D as an operations center, not just a chat surface with avatars.

A few of your points map directly to active/planned work:

• Multi-agent visibility / office as ops center
  • this is the core Office Systems direction and the main product bet
• State -> animation mapping via config
  • agreed; this should become schema/config driven rather than hardcoded
• Mobile-first office UX
  • also agreed; we already have mobile work in progress and need the shell/navigation to become intentionally mobile-friendly
• Sound / ambient / event cues
  • this fits well with the office model and we want to treat it as operational feedback, not just polish
• Enterprise auth / reverse proxy / OIDC
  • this is highly relevant, especially after the recent hardening work; documenting oauth2-proxy / Entra-style deployments is on the radar
• External event webhooks
  • strong fit for Claw3D; event ingress into the office world is a very natural next integration surface

We’ve also started a diagnostics lane (claw3doctor) specifically because deployment/runtime issues across OpenClaw, Hermes adapter, tunnels, auth, and custom runtimes were becoming too repetitive to debug manually.

So the short version is: yes, this feedback is directionally aligned with the roadmap, and I’ve folded these requests into the ongoing planning around:

• runtime profiles
• Office Systems
• mobile shell
• state-to-animation schema
• event ingress / webhook reactions
• enterprise auth/deployment guidance

If you’re open to it, your setup is exactly the kind of real-world usage pattern that would make you a valuable early tester once those pieces start landing.

[iamlukethedev]

@Jbennett111 we are working on all the points you mentioned above.