ModelMachines.ai
legacy

This page is being migrated

The old data source for this view is not tenant-scoped, so it has been hidden while platform_backend grows the tenant-safe equivalent. It will return here once the new contract is available.
In the meantime: use the Workloads page for run lifecycle, the Endpoints page for runtime/billing state, and the Wallet page for credit history.
Dashboard

Dashboard

Snapshot of your workspace — credits, recent workloads, and active endpoints.
Workloads

Workloads

Pipeline runs submitted from this workspace. Each one reserves credits up front, runs asynchronously, and reconciles when complete.
Workload

—

—
Endpoints

Endpoints

Runtime state of deployed serving endpoints — class, hourly rate, status, and grace expiry.
Wallet

Wallet

Credit balances and transaction history for this workspace.

Transaction history

Audit

Audit log

Every meaningful action taken in this workspace — workload submissions, credit operations, endpoint state changes, and runtime ticks.
Internal admin / Tenants

Tenants

Look up a tenant by ID and act on it. The control-plane API does not currently expose a list-all endpoint, so this is a search-and-act surface.
Internal admin scope. Actions on this page change real platform state for the named tenant — credits, subscriptions, runtime status. No undo from the UI.
Internal admin / Subscriptions

Activate subscription

Provision a subscription for a tenant. Use this to bring a new tenant out of pending_subscription state.
Internal admin scope. Submitting this form provisions credits and changes the tenant's onboarding state. Reversible only via further admin actions.

Subscription parameters

All fields are required unless marked optional.
Free-form string. Must match a plan recognized by the backend.
Credits added to the tenant's wallet on each monthly replenishment.
Move tenant.status from pending_subscription to active.
Issue the first month's credits immediately, rather than waiting for the next replenishment cycle.
Internal admin / Credits

Credit operations

Manual adjustments and out-of-cycle monthly replenishments. Two separate actions on the same screen because they share a tenant context.
Internal admin scope. Both actions create real wallet transactions visible in the tenant's audit log.

Manual adjustment or top-up

Use adjustment for corrections (positive or negative). Use top_up for paid additions.
For adjustment, negative values are allowed (e.g. -10 to deduct 10 credits). top_up requires a positive value.
Persisted on the transaction for audit.

Monthly replenishment

Issue the configured monthly credit amount. Override the amount only if you need to issue a custom value for this cycle.
Leave blank to use the tenant's plan-configured monthly credit amount.
Internal admin / Reconcile

Workload reconciliation

Set the actual credit cost for a completed workload. The estimate held during submission is reconciled against this value — refund or charge the difference.
Currently a manual flow. Per the integration contract, reconciliation is operator-input today rather than auto-derived from cloud billing. Use the actual_credits value you've validated externally.

Reconciliation parameters

The workload must already be in a terminal state (succeeded/failed/cancelled).
If higher than the estimate, the difference is charged. If lower, the difference is refunded.
Internal admin / Global audit

Global audit log

Every audit event across all tenants. Same vocabulary as the tenant audit page, no tenant scoping.
Internal admin scope. This view spans all tenants. Treat tenant_id values as identifying.
Internal admin / Runtime

Runtime ops

Manually trigger the runtime metering tick. In production this runs on a schedule; this control is for diagnostic and recovery use.
Per the integration contract, the runtime meter is currently API-triggered rather than scheduler-driven. Manual ticks consume credits across all currently-active endpoints.

Trigger metering tick

A tick walks all active endpoints, computes credits consumed since the last tick, and writes the resulting transactions. Side effects: wallet balances change, endpoints may transition to depleted_grace or suspended.

New workload

Paste a blob URL to an .ipynb or .py file. Credits will be reserved up front based on an estimate.
Launch

Run a pipeline

Point to a notebook on GitHub and the pipeline will instrument it, build a training image, train on the cluster, register the resulting model, and deploy it. You'll see progress live as each stage runs.
Paste a blob URL to an .ipynb or .py file. The pipeline fetches it, instruments it for tracking, and runs it through the cluster.
Deployments

Live endpoints

Serving endpoints with an attached deployed model. Each row is one serving surface — if an endpoint splits traffic across multiple models, each split is listed separately.
⌘> ▏
APIs

Prediction APIs

Callable prediction endpoints for models live on serving infrastructure. One row per deployed model — click any row for the curl-ready request.
⌘> ▏
Model registry

Versioned models

Versioned models in this workspace. Click a model to inspect versions, aliases, and deployed endpoints.
⌘> ▏
Training

Training runs

Experiment runs that produced the models in this workspace. Each row is one experiment from an upstream pipeline run, with the parameters, metrics, and model versions it produced.
⌘> ▏
Artifacts

Run artifacts

Files captured during training — confusion matrices, error plots, JSON outputs. Grouped by training run, then by the model each artifact belongs to. Previews load on demand.
⌘> ▏
Container registry

Container images

Container images available to this workspace. Digest is the stable identity — tags are mutable and point at a digest.
⌘> ▏

Backend configuration

Address of the API serving this workspace. The backend must be reachable from your browser, with cross-origin access allowed for this page.