Use casesModel changes
Models change under you. Find out before your users do.
Vendors retire models, swap snapshots and change defaults on their own schedule. INFRO warns you while you still send a retiring model, names the replacement, lets a route move traffic without a deploy, and reports changes in your traffic that come with no error.
Example finding. INFRO sees request metadata, not answers, so it does not judge answer quality.
The problem
Without a control plane
- A retired model fails in production on the vendor's shutdown date.
- A new snapshot answers longer, slower or without the cache, silently.
- Moving to a new model is a code change in every service.
With INFRO
- Retiring models flagged while you still send them, with the date.
- Traffic changes reported with their likely cause and weekly cost effect.
- A route name keeps working while its target changes.
How it works
Get warned
INFRO records each vendor's deprecations and shutdown dates and warns organizations still sending a retiring model.
Move with a route
Point a route that carries the old name at the replacement. Route names resolve before catalog models, so callers do not change.
Watch what changed
Traffic changes flag a new snapshot, longer answers, a lost prompt cache, more fallbacks or slower responses, each with its likely cause.
What you get
Each one is documented, and each says which plan includes it. Where no plan is named, it is on every plan.
Retirement warnings
Organizations still sending a retiring model are warned ahead of its shutdown date.
DocsRefused by name, never swapped
From the shutdown date the model is refused before any call with model_shut_down, naming the replacement.
DocsRoutes as stable names
A route can carry the old model's name and point at its replacement, so no caller changes.
DocsTraffic changes
Changes with no error: a new snapshot, longer answers, a lost prompt cache, more fallbacks, cut-off answers or slower responses.
DocsAvailability from your keys
Your own key's model list is read every three hours, and a model it stops listing stops being offered.
DocsFindings where you work
Findings by email on every plan, and to Slack, Teams, PagerDuty or a signed webhook where alert channels are included.
Docs
Who it is for
- Platform teamsA vendor's deprecation notice becomes a finding with your traffic attached, not a calendar reminder.
- On-call engineersA refusal that names the replacement instead of a generic 404 at 00:00 UTC.
- Engineering managersA cost change from a vendor's snapshot is caught in days, not on next month's invoice.
FAQ
Will INFRO move my traffic to the replacement automatically?
No. INFRO never answers with a different model on your behalf. It refuses a shut-down model by name, and you decide whether a route points at the replacement.
Does INFRO judge whether a new snapshot answers better or worse?
No. It sees request metadata, not answers, so it reports measurable changes (length, latency, cache, fallbacks, errors) and their estimated cost effect, not quality.
Where do the retirement dates come from?
From each vendor's own deprecation pages, recorded with the date they were read. The public AI updates timeline lists the same releases and shutdowns.
The full contract is in the Traffic changes docs, and every other question is on the FAQ.
Field guides
- Field guideSeptember 3, 2026: provider incidents and failoverOn 3 September 2026 three AI providers posted incidents within hours of each other. Their own status pages tell a more precise story than the headlines, and it is the story failover has to be designed for.
- Field guideFailover drills: test LLM provider failure before it happensThe failover path is the least-exercised code in most AI applications, and it runs for the first time during an outage. A method for drilling it on purpose: which faults to inject, a mock upstream that injects them, and the assertions that prove recovery happened.
Get started
Try it on your own provider accounts.
Connect a provider, create a gateway key and change one base URL. Your providers keep billing you directly; 0% markup on model spend.