execution.nativeIn development
Native model execution
Model execution inside the scheduler — local inference paired with frontier-on-demand routing.
Research
Engineering the runtime.
The runtime is shipping. The platform comes next. The AI OS is the long arc. These are the research investments behind each.
$kx --research::5 in-dev·2 planned·1 long horizon
Research areas05
delivery.exactly-onceIn development
Exactly-once delivery
Exactly-once task semantics with transparent recovery.
resource.allocationIn development
Resource-efficient orchestration
More throughput per resource-unit through bin-packing and shared loading — not overprovisioning.
observability.fullIn development
Full observability
Every task state transition as a typed event. Nothing changes silently.
policy.layerIn development
Policy & safety layer
Safety, privacy, and policy encoded as runtime invariants. Non-override by design.
Platform research02
scheduling.adaptivePlanned · platform
Self-improving scheduling
Scheduling that improves with telemetry — bounded and observable, not model weights.
workflow.multi-agentPlanned · platform
Multi-agent workflow orchestration
Composing multi-agent workflows into executable task DAGs.
Long horizon01
ai.osLong horizon
Cognitive AI Operating System
The north-star endgame — revisited after the runtime earns validation. Long-horizon, not a near-term deliverable.
Early access
Be first as the runtime ships.
Join the waitlist for access to the runtime and the platform as each phase reaches early users.
Join the waitlist