Platform feature · Pre-execution planning
Preflight: decide before launch.
Before a job launches, Preflight lays out execution options across reachable providers and chips with predicted time, cost, energy, and accuracy.
How it works
Inspect workload
Read compute blocks, branch structure, and cross-stage transfer volume before allocating nodes.
Score reachable pools
Evaluate chip combinations across the cloud accounts, HPC clusters, and data centres your team can reach.
Compare four metrics
Review predicted time, cost, energy, and accuracy side by side for every candidate option.
Approve & dispatch
Select by your goal—or let an agent policy pick—with nothing spent until selection.
What you get
- Eliminates blind cluster allocations and surprise bills from unpriced retries.
- Matches each run explicitly to speed, cost, energy, or accuracy priorities.
- Gives AI agents a pre-execution budget guardrail on every spawned subtask.
When to use it & who it is for
Built for teams evaluating placement trade-offs before running heavy jobs, and AI agent workflows launching dynamic sub-tasks.
Paired directly with Workload Optimization
Preflight uses the same engine partitioning and scoring as Workload Optimization—the plan you pick before launch is the exact plan executed.
Preflight pre-execution options
Compares time, cost, energy, and accuracy across candidate options before any compute node starts.
Preflight · Simulation sweep
ExampleLay out run options before allocating a cluster
Workload input
Submitted: multi-step physics parameter sweep with sparse solves and dense grid kernels.
ORIQX action
Preflight inspects block ratios and scores reachable CPU, GPU, TPU, NPU, and QPU options.
More ORIQX platform features
Explore the rest of the ORIQX platform.
Workload Optimization
Takes your existing program unchanged, splits the work into compute blocks, and runs each block in parallel on the chip that suits it, fully managed or self-managed.
Explore OptimizeEstimate
Inspects the work your program performs and scores options across chip classes—with zero access to candidate hardware and nothing executed.
Explore EstimateAgent controls
Sits beneath existing agent frameworks so spawned retries, searches, and child simulations pass Preflight cost, speed, energy, and accuracy checks before nodes allocate.
See agentic compute