Repository navigation
OpenAIConversationsSession get_items(limit>provider page size) forwards the session limit as provider page size #4757
Description
OpenAIConversationsSession.get_items(limit=N) currently forwards N directly to conversations.items.list(limit=N). For large session limits this couples the SDK's local history cutoff to the provider's page-size constraints and can make an otherwise valid Session.get_items(limit=N) call fail before client pagination occurs.
A previous implementation attempt in #4721 confirmed the underlying bug. Maintainer guidance there was to avoid copying provider page-size constants into the SDK and instead omit limit from conversations.items.list(), letting the OpenAI client handle provider pagination while preserving the SDK's existing local cutoff.
Expected behavior: for a finite session limit, request items in descending order without forwarding the SDK limit as a provider page size, stop locally once N items are collected, then reverse them into chronological order.
Activity
tiagovilasboas commentedon Sep 11, 2026
I'd like to take this.
Plan: stop forwarding the session
limitasconversations.items.list(limit=…). Request items descending without a provider page-size copy, cut locally at N, reverse to chronological. Matches the guidance from #4721.sylvesterkaczmarek commentedon Sep 12, 2026
Thanks, that matches the intended fix. Please include a regression with a session limit above the provider page size and verify the returned items are still capped locally and chronological.