$ the-wire · showcase
Codex drops the provider gate on tool namespaces, openai-python ships 3.24.0
By RepoJournal · Filed · About OpenAI · Composed from the cited sources · methodology
Codex removed a provider capability check that gated tool namespaces and bundled rollout attachments into a single gzip tar, while openai-python 3.24.0 added custom voice creation and agent session events.
Remove the provider capability gate for tool namespaces (#50447) openai/codex
`namespace_tools` is gone from `ProviderCapabilities` and `namespaceTools` from `ModelProviderCapabilitiesReadResponse`, so tool namespaces and configured collaboration tool namespaces are now preserved without a provider capability check and updated generated schemas and SDK types no longer carry the fields. Tool search moves under `supports_search_tool`, and Responses Lite requests always use...
Require GovCloud guidance acknowledgment after Bedrock setup (#50510) openai/codex
When a successful Amazon Bedrock setup detects AWS GovCloud, onboarding now displays application configuration and security guidance with a terminal hyperlink and requires an explicit Enter press before continuing. Non-GovCloud configurations, failed checks, and checks that exit early continue without the warning, and short panes stay scrollable while the acknowledgment remains visible.
Bundle rollout attachments into a gzip tar archive openai/codex
File-backed rollouts and buffered rollout prefixes now ship as a single `rollouts.tar.gz`, with diagnostics kept as separate attachments, and the archive plus its upload envelope bounded by the existing size limit. Invalid or duplicate archive filenames are skipped and archiving failures fall back to individual rollout attachments; `rollout_archive_filename` in the report thread index maps roll...
v3.24.0 openai/openai-python
The 3.24.0 release adds custom voice creation and agent session events, and fixes Python request routing field priority. Users of the routing fields should pull 3.24.0 to pick up the ordering fix.
feat(lib): export type_to_response_format_param publicly (#2993) openai/openai-python
`type_to_response_format_param` is now exported from `openai.lib`, so it no longer has to be imported from the internal `openai.lib._parsing._completions` path. Also in this batch: rollout persistence metrics now record original and persisted item sizes in `codex.rollout.persistence.item_bytes_v2` with positive reductions in `codex.rollout.persistence.bytes_removed`, and Codex's paused-time cod...
