AI Reliability Control Center

Keep production AI aligned with customer-defined rules, workflows, and routed-call budgets.

Agent QA Status
—
Tool Gate
—
Action Gate
—
Budget Gate
—
Deploy Gate
—
Doctor
—

Budget enforcement applies to configured provider calls routed through the AI Reliability gate. Direct provider calls outside the gate are not in enforcement scope.

How to read this dashboard

  • Healthy means the latest maintenance run matched the active eval spec for outputs, tool calls, actions, and routed-call budget rules.
  • At risk or failed means one or more checks deviated from the customer-defined source of truth, including budget blocks on routed calls.
  • Deploy gate can fail CI/CD when blocking failures or misconfiguration are present (npm run gate:release).

Production Health

Status
—
Latest maintenance run
—
Workflow
—
Environment
—
Spec version
—
Run reference
—
Health signal
—

Doctor

Source-of-Truth Specs

Customers define policies, expected behavior, required tool calls, forbidden actions, and budget limits as eval specs. AI Reliability compares maintenance observations against those definitions.

Active spec
—
Workflow covered
—
Policies recorded
—
Output and behavior checks
—
Required tool-call checks
—
Forbidden action checks
—

Maintenance Gate Results

Checks run
—
Passed
—
Failed
—
Blocking failures
—
Warnings
—

Budget Gate

Monthly routed-call budget
—
Tracked routed-call spend
—
Remaining routed-call budget
—
Per-run limit
—
Latest routed-call estimate
—
Budget gate status
—
Estimated blocked spend
—
Routed calls (this run)
—
Fail closed on unknown pricing
—

Spend shown here reflects calls routed through the AI Reliability gate and recorded in the provider-call ledger. It does not represent your full provider account usage.

Provider-Call Ledger

Total entries
—
Completed
—
Blocked
—
Failed attempt
—
Estimated blocked spend
—
Latest provider and model
—
Time Status Provider Model Est. USD Actual USD Block reason

Evidence

Check Severity Reason Type

Evidence detail

    Recommended Actions

      Deploy Gate

      Deploy allowed
      —
      Exit code
      —
      Agent QA gate decision
      —
      Manual review required
      —
      Blocking failures
      —
      Misconfigured count
      —
      Stale artifact warning
      —
      Decision reason
      —

      Deploy Gate uses maintenance and Agent QA artifacts. Manual review is never treated as deploy-safe. Exit codes: 0 allowed (healthy or at-risk without manual review), 1 blocking failures, 2 misconfigured.

      Remediation

        npm run gate:release