Give each device a clear job
Mobile handles conversations, agent switching and notifications. Desktop shows task dependencies, files and live activity.
Quick check-ins and detailed review have interfaces suited to their context.
AI agents · Product design · Development
Multiagent OpenClaw setup
Delegation works better when the work stays visible.
I built a mobile client and desktop monitoring around my OpenClaw setup so I could follow agent work and review results wherever I was.
- Contribution
- System setup, product design & AI-assisted development
- Period
- 2026
- Status
- Working personal setup · Ongoing development
- Working
- personal agent setup
- Mobile + desktop
- conversation and detailed review
- Čistý kout
- first product built with the setup
agYAP / Louise
A voice for the agents.
A voice for the agents.
A view of the work.
A working demonstration of voice chat, agent switching and mobile updates, followed by desktop views of tasks, files and live activity. The pixel-art office is adapted from an open-source project.
- On mobile
- Talk, switch agents, receive updates.
- On desktop
- Follow dependencies, files and live activity.
The challenge
I configured agents for writing, video, coding and testing. I then needed to stay involved away from my desk and understand what they were doing.
Starting a task was only one interaction. Following progress, inspecting output and deciding what needed attention also needed an interface.
My role
I configured the workflows and designed and built a mobile client and desktop monitoring with AI assistance. The agYAP / Louise client connects through a cloud gateway with Czech and English voice and text.
The decisions
Build on existing foundations
I used OpenClaw and adapted an open-source pixel-art office.
My custom work focused on workflows, access and visibility. I did not build the underlying OpenClaw framework.
The result
The personal setup works across mobile and desktop. Čistý kout was the first product built with it, providing a real use beyond the demo. No team-wide productivity improvement is claimed.
What I’d measure next
- Measure
- Accepted task completion, review effort and recovery from failed work.
- How
- Log comparable tasks with and without the setup. Track elapsed time, active review time, revisions and accepted outputs. Report results by task type; keep personal observations separate from team evidence.
- Check alongside it
- Judge output quality alongside speed and count failed attempts, not only successful runs.