$ the-wire · showcase
Live WebRTC standby lands in both SDKs, Codex splits its policy crates
By RepoJournal · Filed · About OpenAI · Composed from the cited sources · methodology
Codex shipped the Live WebRTC standby session API to openai-node 7.62.0 and openai-python 3.57.0 on the same day, while its own execution policy layer was split so config, protocol, and prompt code stop dragging in the Starlark parser.
feat(api): support Live WebRTC standby openai/openai-python
Python 3.57.0 adds Live WebRTC standby: sessions now pause when idle and resume on speech, configured through a speech wake policy, VAD threshold, and idle timeout at session creation. Readers get session.paused and session.resumed as typed sideband events and can read optional active and paused billable seconds in session usage, so billing and pause/resume handling go through typed fields rath...
feat(api): support Live WebRTC standby openai/openai-node
The TypeScript SDK's side of the same feature moves session usage to whole seconds with optional active and paused billable seconds, a shape that will shift code that assumed a single usage field. Landing in 7.62.0 alongside the matching gpt-live-1.1 model entry means Live examples in both SDKs now point at the new model name.
Split in-memory execution policy rules into a separate crate openai/codex
Policy construction, rule validation, and evaluation moved into a new crate, codex-execpolicy-rules, which configuration, protocol, and prompt code now use directly instead of reaching through codex-execpolicy. Starlark policy loading and file updates stay in codex-execpolicy, which re-exports the shared rule types, so downstream crates no longer compile the parser just to hold in-memory rules.
Refresh Guardian policy on config reload and account changes openai/codex
Running threads previously kept whatever approval requirements and Guardian policy they started with; they now refresh after local config layers are published, after login while the account owner is still current, and after logout. The refresh is attempted even when unrelated user configuration fails to load, which matters for anyone whose config file is partially broken.
Refresh plugin recommendations after installed state changes openai/codex
Recommendation lookups no longer poll the plugin service for a tooling-state marker; recommendations refresh on the next lookup after a known install event or a change to the installed plugin inventory or enabled state. Also worth a look: the Windows runner now explains ERROR_LOGON_TYPE_NOT_GRANTED (1385) with the "Allow log on locally" and "Deny log on locally" settings instead of just an erro...