OpenAI is shutting down its reusable Prompt Objects API (v1/prompts) on November 30, 2026. OpenAI's official recommendation is to move prompts back into version-controlled code — which means every prompt edit becomes a redeploy. Promptary keeps the workflow you actually want: every prompt gets a stable REST endpoint, so editing a prompt updates your app instantly with no rebuild required.
OpenAI's Prompt Objects let you store a prompt template centrally, give it an ID and version, and reference it from API calls. Your production code called a prompt by ID; your team edited the prompt in the dashboard; the change went live on the next request. No rebuild, no redeploy.
That entire workflow ends November 30, 2026:
OpenAI's official migration guidance is to move prompts back into your codebase: keep each prompt in a version-controlled helper file, replace prompt variables with typed function parameters, and manage changes through git history, PR review, and release tags.
That's a legitimate engineering pattern. But developers in OpenAI's own community forum immediately identified what it gives up:
There's a third option OpenAI's guidance doesn't mention: keep prompts in a dedicated prompt management tool with its own REST API — one that isn't coupled to a single model provider's roadmap.
That's the model Promptary is built on. Every prompt gets a stable UUID and its own endpoint:
curl https://promptary.dev/api/v1/prompts/YOUR_PROMPT_ID \ -H "Authorization: Bearer pk_live_..."
{
"id": "...",
"title": "Support reply generator",
"framework": "RACE",
"text": "...assembled prompt text...",
"topic_id": "...",
"updated_at": "2026-07-08T..."
}
Your application fetches the prompt at runtime. Your team edits it in a visual workspace. The next API call returns the updated text. No redeploy, no rebuild, no vendor lock-in — the same prompt serves OpenAI, Anthropic, Gemini, Mistral, or a local model, because the prompt store is independent of the model provider.
GET /api/v1/prompts/:id call to Promptary, then pass the returned text to the Responses API (or any other LLM API) as usual.Prompt versioning is built in on the Team plan — every edit to a workspace prompt is snapshotted automatically, with one-click restore. That covers the "review, compare, and roll back" part of OpenAI's guidance without building your own versioning layer.
If you need hosted eval running (the other November 30 casualty), pair Promptary with a dedicated eval framework like Promptfoo — prompt management and evaluation are different jobs, and tools that do one job well beat platforms that get deprecated together.
Personal plan is free forever, no credit card required. Every prompt gets a REST API endpoint on the Developer plan — no redeploy needed to update a prompt again.
Start for FreeNovember 30, 2026. Prompt creation was de-emphasized starting June 3, 2026, and the Evals Platform goes read-only on October 31, 2026. After November 30, API requests referencing OpenAI prompt IDs will fail.
OpenAI's official guidance is to store prompts as version-controlled files in your codebase and manage changes through git, PR review, and release tags. This works, but turns every prompt edit into a code deployment.
Yes — that's the core workflow Promptary preserves. Each prompt has a stable REST endpoint (GET /api/v1/prompts/:id). Edit the prompt in the Promptary workspace and your application receives the updated text on its next fetch, with no rebuild or redeploy.
Yes. Promptary stores and serves prompt text over a plain REST API, independent of any model provider. The same prompt can be sent to OpenAI, Anthropic Claude, Google Gemini, Mistral, or a self-hosted model.
Yes, on the Team plan. Every edit to a workspace prompt automatically creates a version snapshot with one-click restore — no manual version management required.