$ the-wire · showcase
Vault metadata filters land in both SDKs, shida joins Live voices
By RepoJournal · Filed · About OpenAI · Composed from the cited sources · methodology
2 people shipped this
The vault and Decisions API surface expanded in tandem across openai-node and openai-python, while Codex spent the day hardening the TUI and app-server request plumbing.
feat(api): add vault updates and metadata filters openai/openai-python
Vaults can now be renamed or have their metadata replaced outright, and vault and credential listings accept metadata key-value filters, in both sync and async clients. Empty-string filter values are preserved rather than dropped, while None is still omitted, so a filter of "" now selects instead of silently vanishing.
feat(api): add vault updates and metadata filters openai/openai-node
The Node SDK gains the same typed vault update parameters and metadata filters for vault and credential lists. Teams running both SDKs against one backend can now filter consistently from either side.
Add connection-bound reply handles for app-server requests openai/codex
Server request IDs can be reused across connections, so retained replies now capture the originating client and request ID in an `AppServerReplyHandle` exposed via `AppServerRequestHandle::server_request_reply_handle`. Resolution and rejection target the connection that emitted the request even after the UI selection moves on.
Reuse and prewarm the WSL clipboard text reader openai/codex
WSL clipboard reads used to spawn PowerShell on every paste, and process creation under WSL can hang and stall the shared clipboard worker. Codex now keeps a persistent reader on a dedicated thread, restarts it after errors, and prewarms it at startup when right-click paste is enabled.
feat(api): add shida to named Live voices (#4053) openai/openai-python
shida is now a named built-in Live voice in both SDKs, with existing voice names and open-string inputs unchanged. The 128-character safety identifier limit is also now documented, and the Decisions docs spell out predicate, choice, and score questions plus the 128-image input cap.
