OpenAI shut down the Assistants API on August 26, 2026

We counted how many public n8n templates still build on it

On September 3, 2026 we read every public template in the n8n library that calls OpenAI — 1,362 templates — and checked which ones reach the Assistants API, either through the OpenAI node's assistant resource or through a raw call to api.openai.com/v1/assistants.

19 of them still do. These are not obscure. The most-viewed has over 65,000 views, and eight of them have more than 10,000:

A view count is not an install count, and a template being broken in the library does not mean any particular copy of it is broken in production. What it does mean: these patterns were copied into real client workflows, by real agencies, over the last two years — and nothing about copying a template tells you when its foundation gets retired.

Why this one is easy to miss

1. 01

   It is not a model swap, it is a whole API

   Model retirements at least produce a recognisable error naming the model. An Assistants workflow fails at the API surface — thread creation, run polling, file retrieval — and the error text does not say "this product was discontinued".

2. 02

   Assistants workflows are usually the quiet ones

   Draft replies, meeting summaries, RFP processing, retrieval over uploaded files. They run in the background and produce something a human reads later. When they stop, nothing alarms — the output just stops arriving.

3. 03

   The migration is not a one-line change

   Assistants carried state on OpenAI's side: threads, runs, attached files. The Responses API and Conversations API split that differently, so the workflow has to be rebuilt rather than repointed.

Client-visible symptom: the daily digest stops, the draft replies stop appearing in Gmail, the meeting notes go missing — noticed by the client, usually days later, and usually described as "the automation broke".

How to check your own client workflows today

1. 01

   Export each client workflow and drop it into the free dependency report — it flags the Assistants API surface along with every other announced deadline that applies. The file is parsed in your browser and never uploaded.

2. 02

   Or search the export by hand for "resource": "assistant" inside OpenAI nodes, and for api.openai.com/v1/assistants inside HTTP Request nodes. Both are the same failure.

3. 03

   Rebuild what you find on the Responses API, with Conversations where the workflow needs to remember earlier turns. Run it end to end before telling the client it is fixed — the output shape changes, and downstream steps that parse it will need checking too.

4. 04

   While you are in there: gpt-4, gpt-3.5-turbo, o1 and 26 other model ids shut down on October 23. In the same 1,362 templates we found 130 pinned to a retiring model id. Same audit, two deadlines.