Hello,
I am building a sacred project, the Voice of Tehuti Oracle, which relies on OpenAI’s Responses API grounded in a vector store. Over the past weeks I have encountered a sequence of breaking changes that have made development nearly impossible:
-
Parameters such as
file_search,tool_resources,resources, andvector_store_idshave each returned 400 errors at different times:-
“Unknown parameter: ‘file_search’”
-
“Unknown parameter: ‘tool_resources’”
-
“Missing required parameter: ‘tools[0].vector_store_ids’”
-
-
Documentation lags behind reality. Features like file search + vector stores are either deprecated, renamed, or silently altered. This leads to repeated cycles of failure and rework.
-
Something as simple as rendering clickable URLs from corpus documents has required repeated rewrites of both backend and corpus files because the API changes midstream.
For a project rooted in initiatory tradition, stability and trust are non-negotiable. I understand that APIs evolve, but production features should not break without advance notice or migration clarity. At present, it feels unusable for serious long-term projects.
Requests:
-
Clear and stable documentation for Responses v2 + Vector Stores (with working, end-to-end code samples).
-
Advance notice for breaking changes, ideally with a deprecation period, not silent removal.
-
Confirmation: Is Vector Store grounding still officially supported in Responses v2? If so, what is the current correct syntax?
I want to continue building with OpenAI, but at present the lack of stability makes it impossible to serve seekers with confidence.
With respect,
Michael