Skip to content

Model & Compute — UI surfaces

Target

This page describes intended behaviour or structure. No implementation is claimed unless a section is explicitly marked Current implementation or Evidence.

Screens fed by the platform

Studio work area Screen Read model Commands
Intelligence Model profiles: capabilities, privacy class, limits, fallback chain, allowed deployments GET /factory/model-profiles, GET /factory/model-profiles/{profileId} none (profiles are published through Assets)
Intelligence Deployments: provider, region, state, health, license metadata, capability scores GET /factory/model-deployments RegisterDeployment, EvaluateDeployment, ApproveDeployment (approval-gated)
Intelligence Route-plan preview for a scope and profile with exclusion reasons POST /factory/model-profiles/{profileId}/route-plan/preview none
Operations & usage Run detail → model receipts (deployment, outcome, usage status, tokens, cost estimate) GET /factory/runs/{runId}/model-receipts none
Operations & usage Compute pools: kind, region, boundary, utilization, queue, cached models; placement decisions GET /factory/compute-pools, GET /factory/compute-pools/placements/{placementId} RegisterComputePool, DrainComputePool, WarmModel

State families shown

Family Values
Deployment Candidate, Evaluated, Approved, Degraded, Retired
Usage status Reported, Estimated, Unknown
Invocation outcome Completed, Partial, Failed, Denied, Cancelled
Pool Draft, Active, Draining, Retired
Placement Placed, NoCapacity, PolicyDenied, Deferred

Rules the clients must honour

  • Profiles, never model names, in every picker that an agent, workflow or project policy uses; provider model names appear only on deployments and receipts.
  • Evidence is visible: a deployment's capability scores and license metadata are shown before the approve action, which is an approval-gated publication-class action.
  • Usage is honest: Unknown is labelled, never summed into a final total.
  • Exclusion reasons are shown, not hidden, in route-plan previews and placement decisions.

CLI and VS Code

factory models profiles|deployments|preview-route, factory compute pools|placements; the VS Code run panel shows the model receipts of the bound run.