Shutdown audit
$19924 h
Reserve an audit- Every call, ranked by risk
- Credited toward a migration
Scan your code free for every model OpenAI is retiring. I move each call to its replacement and send a tested pull request within 48 hours.
After each date, calls to the old model fail with model_not_found. The list comes from OpenAI's own page.
Paste code, a .env or an n8n, Make or Zapier export, or drop a folder. Each match shows its file, line, date and replacement. Nothing leaves your browser.
Results show here: the model, where it is, the day it stops and what replaces it.
Give sunsetfix-bot read access, or send the n8n, Make or Zapier export. I never ask for passwords or API keys.
Every call goes to its replacement with the settings it needs. Your saved inputs run on the old and the new model.
The report rides along: what changed, which answers differ and the cost per call before and after. Seven days of fixes included.
Fix it once
$19924 h
Reserve an audit$54948 h
Reserve a migration$99048 h
Reserve a rescue$1,4905 projects
Reserve agency slotsSunsetFix checks your repos against OpenAI's deprecations list and opens a tested migration pull request before each date.
Early access: I set it up by hand for the first teams.
sunsetfix/gpt-5-miniwatch14 minRunning#215deprecations page1 min3 repos2 mingpt-5.6-terra4 min20 saved inputsQueued+ Slack alertQueuedScansEvery week, plus one within 24 hours of any OpenAI notice.
FixesA tested pull request for every flagged call, within 48 hours of the notice.
AlertsBy email, or in Slack on Watch plus and Agency.
Watch plans: monthly, cancel any time
$199a month
Join Watch early$349a month
Join Watch plus early$499a month
Join Agency earlyWho does the work: Riccardo, an independent developer in Rome. I write the code with Claude and test every change before you see it. Replies the same day.
Often. The replacements handle reasoning, token limits and output format differently, so prompts can drift. The report shows whether they did.
No. You run the compare script with your own key, or add a temporary key to a test project you control.
OpenAI's export gives Agents SDK code but does not convert the graph. I rebuild it, keep your ChatKit widget and test the answers.
Not yet. In early access I connect it to your repos by hand, with read access and a bot account for the pull requests.
After you approve the written scope. Late delivery or a failed test case means a full refund.