Skip to content

Search docs

Find pages, headings, and concepts. Press ⌘K or Ctrl+K to toggle.

Admin operations

Platform administration for governance, telemetry, support workflows, and operational control.

Admin mode is available at /admin for platform operators. The surface is intentionally separate from normal org navigation and requires admin-only access to protect tenant-level controls.

What you manage in admin

  • Identity and ownership — orgs, users, workspaces, and org-scoped role boundaries.
  • Usage and spend control — billing health, usage exports, and operational spend review.
  • Connector governance — connector catalog health, connection state, and runtime webhook delivery.
  • Platform quality signals — feedback triage, notification queues, admin notifications.
  • Model and commerce health — model catalog and shopping store/product/order snapshots.
  • Trust and safety — security events, runtime health checks, sandbox status, and audit trails.

Admin feedback and triage

/admin/feedback provides two operational modes:

  1. Feedback mode — review user reports and security probes captured by the product.
  2. Notifications mode — monitor platform-level alerts (schedule completion/failures, message events, usage events).

Feedback mode also exposes:

  • security view and classification groups (script, command, SQL, template, path, fuzz)
  • signal quality categories (product_signal, short_actionable, actionable, security_probe)
  • burst/duplicate group hints so repeated probes are triaged once.

You can update workflow state from open to triaged, in_progress, done, then archive once resolved.

Wallets and financial operations

Admin operators can inspect agent wallet operations through dedicated screens and /api/v1/admin/agent-wallet. The page surfaces:

  • Stripe issuing cards and card status
  • spend requests and spend approvals
  • authorization outcomes and transactions

These views help operators investigate card disputes and onboarding/usage anomalies.

Connector event operations

Connector event delivery, event IDs, and verification posture are visible from the admin pages. If an upstream webhook drops retries or a worker stalls, operators use the admin connector-events tools and sandbox/job status views to keep the route healthy.

Destructive super-admin actions

Cross-tenant destructive mutations (refunds, broadcasts, BYOK revoke, connector-event replay, session cancel, newsletter export, CMS publish mutations, skill deletes, and similar) require super_admin plus fresh step-up via requireFreshSuperAdmin. Read-only admin routes accept admin or super_admin without step-up.

Site CMS (marketing pages)

/admin/site/pages edits the public marketing site (site_pages + version history). Reads require admin; create, update, archive, publish, and revert require fresh super-admin step-up and write admin_audit_logs rows (site.created, site.updated, site.published, site.archived, …).

Publishing snapshots the current row into site_page_versions, sets status=published, and revalidates the page path plus /sitemap.xml via lib/site-cms/revalidate.ts. Public rendering only serves published slugs when the global feature flag site.cms_enabled is on.

Deployment targets

/admin/deployments lists deployment_targets and custom domains: routing mode, isolation mode, health, migration version, DNS verification state, and default hosts. Enterprise tenant hosts (*.enterprise.alumia.com or customer aliases) map through this registry — see the Enterprise guide.

AI loop runtime

/admin/runtime exposes resolved rolloutControls for shared agent-loop flags (ai.loop.*). Toggle kill-switch or pruning flags here before changing loop code in production.

Admin console UI strings are localized through navigation message keys. Enterprise prospects use the public Enterprise guide and POST /api/v1/enterprise-inquiry.

For every major rollout:

  • confirm /admin/models and /admin/usage remain responsive
  • run a quick pass on recent /admin/connector-events delivery failures
  • review new open feedback with highest severity and mark actionable handoffs quickly
  • clear queue spikes in /admin/runtime or sandbox views before changing connector/webhook configurations
  • verify enterprise deployment targets show healthy before DNS cutover