Provider Settings
The Providers admin page lists all 8 registered providers with KPIs and a per-provider toggle. Disabling a provider here makes it reject every request gateway-wide, regardless of which project / endpoint / wrapper made the call.
Use this when you want to stop using a provider in seconds without touching credentials or endpoints — during incidents, billing pauses, compliance holds.
What you see
Section titled “What you see”Top-right user menu → AI Providers.
| Element | Meaning |
|---|---|
| KPI strip | Total / Enabled / Disabled / Requests last 24h |
| Card per provider | Identity, status dot, ON/OFF chip, Configure / Disable buttons |
| Disabled cards | Render at 55% opacity — easy to spot at a glance |
Each card also shows:
- MODELS — distinct model identifiers in active endpoints + templates for this provider
- p95 — latency over the last 24h (only
okrequests counted) - SHARE — % of total gateway traffic that hit this provider
Enabling / disabling
Section titled “Enabling / disabling”Click Disable on an enabled card → the provider is now rejecting requests:
- Existing endpoints / wrapper routes return
503 Provider 'openai' is disabled in this gateway. - The card flips to OFF, opacity drops, the action button becomes Enable
Click Enable on a disabled card → restored.
This is implemented in provider_settings (one row per provider key, with is_enabled boolean). Default behaviour when no row exists: enabled. So a fresh install has no rows and every provider is enabled.
Why disable?
Section titled “Why disable?”- Incident — provider is having an outage, you want to fail fast and force failover paths instead of timing out per request.
- Cost — billing alert, you want to stop spending while you investigate.
- Compliance — a region is added to a data-locality requirement; pause the provider until you’ve migrated.
- Testing — verify that endpoint failover chains actually fail over.
What it doesn’t do
Section titled “What it doesn’t do”- ❌ It doesn’t disable individual models (no per-model granularity in v1).
- ❌ It doesn’t preserve in-flight requests (any active call completes, but new ones get 503).
- ❌ It doesn’t disable upstream OAuth connections in the API Gateway — those are independent.
- ❌ It doesn’t notify clients pre-emptively. Clients see 503 on the next request.
Detail page
Section titled “Detail page”Clicking the Configure button on a card opens the provider detail page with:
- All credentials registered for this provider
- All provider templates referencing it
- All AI endpoints currently routing through it (so you can see the blast radius)
Useful before you flip the switch.
Each toggle writes an audit entry: provider.enabled or provider.disabled with the provider label as target. Filter the audit log to see who flipped what when.
Programmatic toggle
Section titled “Programmatic toggle”There’s no public API for this — it’s an admin-only UI action. If you need automation (e.g. an incident runbook), use the MCP Control Plane to script it once that’s wired (currently the Control Plane has read tools for providers; toggle endpoints via Tinker).
Next: Adding a Provider — for plugin / extension authors.
© Akyros Labs LLC. All rights reserved.