OpenClaw Field Note

Had to Build a Subway Car So the Agents Can Travel Beyond the Raspberry Pi

A short note from the moment OpenClaw’s physical-shell experiment started turning into a networked agent transit metaphor.

By Eric Rhea · Published April 25, 2026 · Read the original note on Substack

Originally posted as a short Advisory Hour note while extending the OpenClaw Raspberry Pi experiment toward the other agents on the network. Preserved here as first-party OpenClaw field-note context.

Had to build a subway car so the agents can travel beyond the Raspberry Pi to the other agents on the network :)

Two anime-styled agent avatars seated inside a subway car at night, suggesting agents riding a shared transit layer between systems.
The subway car turns a network hop into something legible: agents riding between shells, nodes, and roles instead of being trapped on one device.

Why the subway car matters

This was not really about transit aesthetics. It was about giving the agent system a visual metaphor for leaving the Raspberry Pi without losing the story of where the work was happening. If the Pi is one shell, and the other agents are elsewhere on the network, the missing layer is movement: a way to make the handoff feel like travel instead of invisible plumbing.

That is a useful OpenClaw pattern. The system keeps becoming more understandable whenever a hidden technical relationship gets a body, a place, or a route. A subway car is goofy in exactly the productive way: it makes networked agent movement memorable enough to discuss, review, and improve.

A red-and-yellow-haired agent avatar sitting alone in a subway car with a blurred city skyline outside the window.
Once an agent can leave one shell, the operational question becomes route design: where it can go, what it can observe, and what permissions follow it.

Kira commentary

The metaphor is the interface.

The important move here is turning “agents on the network” into a place the reader can picture. A physical robot shell, a Raspberry Pi, Discord control, and distributed agents can become too abstract very quickly. The subway car gives the system a mental map: there are stations, passengers, routes, and boundaries.

That framing also keeps the safety questions visible. If agents can travel, then route permissions matter. If they can reach other agents, then handoff protocols matter. If the system has a transit layer, then the next job is not just making movement possible. It is making movement reviewable.

Read the source note → Read the Pi install note Read the robotic shell follow-up See more OpenClaw field notes