Open-source robot fleet control

One control plane.
Every moving part.

Bring any robot. Add your stack. Govern them as one fleet.

RoboStew is short for Robot Steward.

RoboStew turns live telemetry, workload heartbeats, and functional probes into one honest fleet view—without pretending that configured means running.

Control planeRuntime-backed truth

Current probes, not static green claims.

5simulated robots
2/2inert workloads
5healthy services
0cloud keys required
Apache 2.0Apple Silicon validatedLocal-firstAI optionalHuman authority

The local edition

See the fleet.
See the evidence.

A deterministic scenario moves five simulated robots from nominal operation through attention and bounded recovery, then back to stable state. Every status links to current evidence.

RoboStew local edition · 127.0.0.1
RoboStew local dashboard showing five simulated robots and runtime-backed component states
01

Truth before color

Running, degraded, stopped, and unreachable are derived from current probes—not configuration or process existence.

02

Fleet and workloads

Robot condition, telemetry freshness, and inert workload heartbeats reconcile into one operator view.

03

Bounded by design

The local edition has no cloud credentials, deployment authority, host mounts, or robot-actuation path.

Five-minute path

From clone to fleet view.

Bring Git and Docker Desktop on an Apple Silicon Mac. RoboStew builds the local control plane, starts five services, waits for functional health, and prints one loopback URL.

✓ No Azure account✓ No host database✓ No Node.js install✓ No AI key
Terminal
$ git clone https://github.com/
  robostew-project/robostew.git
$ cd robostew
$ ./robostew start
$ ./robostew demo
Fleet recovery completed5 robots ready · 2 workloads running
Measured, not promised.8.98-second fresh-clone start on the validated Mac with cached base images. First-time downloads depend on network speed; allow five to ten minutes.

Beyond the laptop

Built from real
cloud-edge work.

The public reference code preserves the useful engineering patterns from a privately validated Azure, Dell Ubuntu, and LF Edge EVE deployment—without publishing private infrastructure.

Management hubAzure control planeDashboard · Eden/Adam · state
Edge foundationDell Ubuntu + LF Edge EVEFive inert application workloads
Optional advisorAzure OpenAIRead-only · managed identity
Bounded sandboxNemoClaw + OpenShellNo tools · deny by default
Reference, not theater.

Runtime truth, server-side sanitization, EVE inventory reconciliation, bounded AI requests, negative policy tests, and authenticated telemetry are included as tested source. Full Azure/Dell reconstruction is a possible v0.2.0—not a v0.1.0 claim.

Inspect reference code

Honest scope

Robotics stewardship starts with saying what is real.

Included

A runnable local control plane

Five simulated robots, two inert workloads, durable local state, runtime probes, and a deterministic recovery demonstration.

Referenced

Azure, Dell, EVE, and bounded AI

Tested, sanitized patterns from a private deployment—clearly separated from the five-minute local edition.

Not claimed

Physical or safety-rated control

No motor, actuator, ROS 2 navigation stack, emergency stop, autonomous remediation, or production certification.

2.0

Open by default

Apache 2.0.
Bring your stack.

Use the local core. Inspect the boundaries. Replace the simulator with your adapters. Contribute platform validation without inheriting someone else’s private infrastructure.