If Core is the rules, Fleet is the hands. It coordinates real compute — GPUs, models and tools — across the portfolio, turning a single decision into a finished, verified result. Every job is described by what it needs, routed to whatever can do it, and tracked from start to finish.
Choose what to run
A lightweight decision engine selects the next job and why — within the bounds Core allows.
Check feasibility
Fleet works out whether the job can run, and where — without committing anything yet.
Claim resources safely
It atomically reserves what it needs with a lease, so two jobs never fight over the same capacity.
Do the real work
A worker runs the job — generating an image, running a model, producing a real asset.
Log it as evidence
The result is recorded with how it was made — measured, not assumed — and the reservation is released.
Capability-based
Work is described by what it needs, not wired to specific hardware. Add a machine and Fleet can use it.
One source of truth
A single authoritative state, so what the system believes always matches what's actually happening.
Pluggable & replaceable
The units that do the work are swappable — nothing is locked to a single vendor, model or tool.
Recorded by default
Every output captures how it was produced, separating measured results from simulated ones.
Fleet runs real AI workloads across real hardware — GPU nodes, including an RTX 5090, generating images, audio and more. The very same pipeline that drafts a product's hero image is the one that scales to production work.