Operator Agent. The execution layer.
CAE plans the workload; the Operator Agent runs it. A Rust agent that executes on the spacecraft, reports power and radiation, pauses on eclipse, resumes on sun, and recovers from single-event upsets - no operator in the loop.
Register a node. It pulls work and reports back.
$ cargo install rotastellar-agent $ rotastellar-agent run \ --agent-id sat-25544 \ --api-url https://runtime.rotastellar.com \ --api-key rs_... ✓ registered · polling for planned workloads # typed events stream back as the workload runs > job.accepted STEP-7F3A > step.completed inference > job.completed
It keeps running when orbit doesn't cooperate.
You can plan workloads in orbit. Who executes them?
A plan is only useful if something on the spacecraft can carry it out - through dropped links, power limits, and bit flips, with no one to intervene. The Operator Agent is that something: a small, memory-safe runtime that turns a CAE plan into executed work and a stream of typed events.
Pull, not push - built for intermittent links.
Ask for work
During a contact window the agent polls for steps CAE has assigned to its node - no inbound connection required, so a dropped link never strands a job.
Run the step
Run a planned step inside the satellite's power and thermal budget, pausing on eclipse and resuming on sun.
Stream typed events
Every action produces a typed event in the same format CAE uses to plan - so execution and planning speak one language.
Heal autonomously
Single-event upsets are nominal, not exceptional. The agent detects and recovers from them without an operator in the loop.
Built in Rust. Tested without hardware.
Put your first workload in orbit.
Tell us what you want to run and the satellite you want to run it on. We'll come back with a feasible plan - or an honest reason it can't be done yet.