Skip to main content
Model Lifecycle watches the models your project actually calls against a curated catalogue of provider-announced deprecation and retirement dates, and raises a finding when one of your models is heading toward end-of-life. Every number on a finding is measured from your own traffic. A model test, when you run it, is a separate and labeled step. With the new agent layout on, the page opens with a card when the project has an open finding. If the retirement alert is off, it is a Decide card with an Alert me before it retires button, shown only if you hold the permission to manage alerts and your plan includes the alert. Pressing it turns the alert on and keeps the email and webhook settings you already saved. If the alert is already on, the card is information only. It names the soonest retirement date it knows and says what it did not check: I do not run your evaluations against it; a model test, when an owner or admin runs one, checks a few of your calls. See Test the successor. Below the plan that includes the alert, the card has no button. If the server refuses the change, the card says what is needed and stays where it is. Turning the alert on from the card does not appear in Activity yet.

What the radar watches

A daily scan (04:00 UTC) sweeps your project’s model usage from the last 30 days and matches every model you called — by model id, disambiguated by provider when the same id is published by more than one — against Zespan’s bundled catalogue of provider deprecation announcements (ModelLifecycle). See the feed reference for exactly what’s in that catalogue and how it’s maintained. A model only produces a finding when both are true:
  • It has an announced retiresAt date, and that date falls within the detection horizon (90 days by default)
  • Your project actually called it at least once in the last 30 days
A model with an announced retirement date that you’ve never called, or one retiring further out than the horizon, produces nothing — there’s no finding to review and no widget to see.
The catalogue is curated and bundled with each release, not scraped from a provider’s page at request time. This also means a self-hosted install with no outbound internet access evaluates lifecycle findings normally — there’s no live fetch in the critical path.

Where findings appear

What a finding contains

Each finding is one (project, model) pair. Everything below is either read directly from the catalogue or measured from your own trace data over the trailing 30 days — nothing is inferred beyond that.

Cost comparison

When the provider named a successor and you have real traffic on it too, the finding shows the measured difference in average cost per call between the two models over the same 30-day window. This is a comparison of your own organic traffic on both models — not a matched, controlled comparison, since the two models may be serving different operations. If you have no traffic on the successor yet, the finding says so plainly instead of showing a blank or a zero.
The cost comparison never says how the successor would answer. To see that, run a model test; see Test the successor below.

Test the successor

When a finding names a successor model, Zespan can test it on your own calls before the old model retires. The panel sits on the finding, under the heading “Test new-model before old-model retires”. It is off when your organization has model tests switched off, and in shadow mode owners and admins see the prepared test with the line “I prepare this test and show it to owners and admins only. Nothing runs until it is switched on.” I prepare a test when a model is within 30 days of retiring. Until then, or after a refused or failed test, an owner or admin can press Prepare a test to read a fresh estimate. Nothing runs until an owner or admin confirms it. While the page is still reading, or the model tests are not installed yet for your install, the panel says so (“Reading…” or “Model tests are not installed yet.”) and does not claim there is no test.
A model test spends on your own model connection, billed to you by your provider, not by Zespan. It runs only after an owner or admin confirms the estimate, and it stops at the limits below. A test through your own endpoint costs whatever that endpoint costs; I can count its calls but cannot price them.
1

Read the estimate

The panel says what I would run, for example “I will make 8 calls on your Prod OpenAI connection, up to $0.02, on 4 of your recent calls." The calls run on your own model connection, so you pay the provider, not Zespan. Only owners and admins see dollars. When I cannot price the calls, for example because they go through your own endpoint, the panel gives the call count and says "I can't see what they cost." It never shows "$0” for a cost it does not know.
2

Run the test

Run the test (owners and admins) reads a fresh estimate first. If the estimate grew since you saw it, or the source changed, I show the new estimate and run nothing. The run stops at today’s test limit and says so.
3

Read the result

The result is a table per model: test runs, checks passed, not checked, errors, median time and, for owners and admins, the cost to you. It says “new-model passed 8 of 8 checks; old-model passed 6 of 8.” It does not claim accuracy, safety or a percentage. A test that stopped early says “This test is partial: I stopped after 2 of 4 items because today’s test limit was reached.” If the old model no longer answers, I test the successor alone and say so.

What the test runs on

I use the first source that has something to test, in this order: If none applies, the panel says I cannot test the successor for that agent and what to add.

Limits

These are the defaults. A test stops at the first one it reaches and says which, as a partial test. Before you confirm, I read the estimate again. If it is older than 10 minutes, grew by more than 5%, or the source changed, I show the new estimate and run nothing. The test writes only its own results. It changes none of your traces, prompts or settings.

Switch to the successor

When Zespan manages the model of the agent, that is, a Zespan prompt version names the retiring model, the result has Switch prompt to new-model for owners and admins. It asks first. Pressing it makes a new version of the prompt on the successor model. Clients that fetch the latest version get it now; clients pinned to a label keep their version until you move the label. I do not move a production or staging label. When I do not manage the model, the panel says “I do not manage the model of this agent. Change it in your code and I will watch the change.” Under the cost comparison the page says either “I test the successor on a few of your calls only when an owner or admin runs the test above. It is not your evaluation suite; read the results before you switch.” (model tests on) or “Zespan does not run your evaluations or regression suite against the successor model. Compare quality yourself before switching.” (off). Someone who is not an owner or admin reads “An owner or admin runs this test. I run nothing until they confirm it.” and sees no button and no dollar figure. On Home and Needs you the same test appears as a card: “old-model is retiring. I can test new-model on 4 of your recent support calls first, on your Prod OpenAI connection, for up to $0.02.” with Run the test, then the result with the switch.

The re-raise ladder and dismissing a finding

A finding’s urgency is expressed as a band: 90, 30, 7, or 0 days remaining (0 meaning the model has already retired). Zespan notifies the first time a finding is created, and again only when the band tightens — the deadline gets meaningfully closer. A daily re-scan that lands in the same band is silent by design: a re-run with nothing new to say would train you to ignore the feature by the time it actually matters. Dismissing a finding (with a required reason) suppresses it at its current band. It stays dismissed through further daily scans as long as the band doesn’t tighten. When the retirement date moves into a nearer threshold — 90 → 30 → 7 days, or crosses into retired — the finding automatically reopens exactly once and notifies again, clearing the old dismissal reason. A band that widens (a feed correction pushes the retirement date further out) never reopens a dismissed finding.
1

Open a finding

From the Model Lifecycle page, the Models page banner, or the Overview widget.
2

Review the evidence

Call volume, cost, affected agents and prompts, and the cost comparison against the successor if you have traffic on it.
3

Dismiss with a reason, or leave it open

Dismissing requires a short explanation (e.g. “migrating next sprint”) so a later reviewer knows why it was suppressed. Leaving it open keeps it visible until you act.

Getting notified

Findings are always recorded and visible on this page regardless of alert configuration — but by default nobody is notified when a new one appears. A small settings card at the top of this page turns that on:
1

Toggle notifications on

“Notify me about model deprecations”, at the top of this page.
2

Add recipients

A comma-separated list of notification emails, and optionally a webhook URL. The webhook is validated against the same SSRF-safety check every other alert webhook in Zespan goes through — a URL that resolves to a private or internal address is rejected.
3

Save

Takes effect on the next daily scan that raises or re-raises a finding — see the re-raise ladder above for when that is.
This is a dedicated, minimal opt-in for this feature only — it does not appear on, and is not part of, the general Alerts page, since a model-lifecycle rule has no metric, condition, or threshold for that page’s table to display. See Alerts → Model lifecycle alerts for how this fits into the shared delivery machinery (same email/webhook pipeline every other alert rule uses).

What’s not built yet

  • No quality score. A model test reports checks passed on a small sample of your calls. It is not an accuracy figure and it does not prove the successor is safe; read the results before you switch.
  • No drift detection. Model Lifecycle only watches for provider-announced retirement dates. It does not detect a provider silently changing a pinned model’s behavior over time — that’s a separate, harder problem tracked for a future release.
  • No automatic migration. I only make a new prompt version when an owner or admin presses the switch. I never change your code, a config file or a label.

Next steps

  • Models — the Lifecycle column and countdown chip on the model registry table
  • Home and Needs you — where a model test card waits for you
  • HTTP targets — test on your own endpoint
  • Alerts — get notified by email or webhook when a new finding is raised
  • Model lifecycle feed reference — the exact catalogue fields, how it’s curated, and how to report a missing or wrong entry