Custom GPT actions intermittently fail to dispatch — "actions not available in this chat session," nothing reaches our server

[Yoav54]

Standard workspace (not FedRAMP), ChatGPT Pro, web.

Our custom GPTs have OAuth-configured actions with a valid OpenAPI 3.1.0 schema. Intermittently, the GPT responds that “the required actions aren’t available in this chat session” and no request reaches our backend (confirmed via server logs — total silence, not an error). Minutes earlier the same GPT, same config, dispatched the same actions successfully. No changes on our side between working and failing runs.

Critically, the Action “Test” button in the GPT builder also fails to fire on these actions — this bypasses the model, instructions, and schema interpretation entirely, so it is not a prompting or configuration issue on our end.

Verified our side is healthy: direct curl to the endpoint returns correct responses (401 without token). The failure is in ChatGPT’s action-dispatch layer, not our server.

Multi-action GPTs are hit hardest; single-action GPTs are more reliable. Intermittent and non-reproducible on identical input.

[iyvljtjqleyj]

Can confirm, but only since yesterday (5.6 SOL). All custom (action) GPT’s are broken since.

My custom GPT's can't retrieve / remember any external data since 5.6 SOL

I have several custom GPT’s for years now, but since yesterday - 5.6 sol - none of them is functional. The workflow is simple. I place an URL in the chat; GPT executes a custom action; It pings my own server; It sends that URL to it; My server fetches data using that URL; GPT receives the data; Done, GPT has the data I need in the chat. This still works, I can see all actions executed in my server, but the GPT can’t use the data since 5.6 sol. It does count the words, does analyze the retrie…

[iyvljtjqleyj]

In my case, the action itself (using a custom action in the web interface) is working.

So the GPT does execute the action, and does fetch data from my external server.

But I only has access to that retrieved data in the first message.

So when I ask a 2nd question about the data (in a follow-up message), it doesn’t remember the data at all. It has to execute the server once again.

---

I also had this exact same problem months ago, when another upgrade / update from OpenAI was forced.

Custom GPT OAuth stuck on “Talking to <domain>” ,looking for official guidance

$_GET works, but in my case it doesn’t remember the data it gets. I can see from both my server-side as in GPT debug environment that it DOES get the data requested. But when I refer to it after the 1st message, it doesn’t remember it and starts hallucinating (it’s not using the data after the 1st message). SCHEME Get data from server Receive data “do something with data” ← this works Get data from server Receive data GPT says : data fetched what’s next “do something with data” ← this fa…

And now it seems the same bug (under different circumstances) has returned.

---

Simply said, this is the situation right now;

OLD SITUATION

1. GPT fetches data from my server
2. I ask question A
3. GPT gives answer A
4. I ask question B
5. GPT gives answer B
6. etc…

---

NEW SITUATION

1. GPT fetches data from my server
2. I ask question A
3. GPT gives answer A
4. I ask question B
5. GPT doesn’t know / have access to the data from step 1
6. I force him to fetch it again
7. GPT gives answer to B
8. I ask question C
9. GPT doesn’t know / have access to the data from step 6 (or 1)
10. I force him to fetch it again
11. GPT gives answer to C
12. etc…

---

The workaround it to directly let the GPT write down ALL RETREIVED data once it receives it.

In that case, the data is available even after some messages.

But sometimes it receives lot’s of data from my server, and can’t write it down all in one message (2.000 words or some limit like that). So in that case it’s summarizing it, and deleting the raw data I needed.