RepoJournal

$ cat openai/week/2026-11-16.log

OpenAI

OpenAI

the week in review · Nov 16 – Nov 22, 2026

SDKs cut alpha budget API, Codex hits 0.170.0

By RepoJournal · composed from the cited sources · human-reviewed weekly · methodology

♥

Python 3.52.0 and Node 7.60.0 both removed Managed Agents session budget types; Node also gained typed Live MCP server configuration.

324 commits 339 PRs merged 32 releases 7 briefings covered

all openai reviews →

feat(api): hide Managed Agents session budget controls and events openai/openai-python

by markstuart-oai

The Python SDK dropped the Managed Agents session budget members from its public types, and the pull request describes it as an alpha breaking change for a minor release. It advises staying on the previous SDK if your code depends on them; the Node package shipped the same removal.

v7.60.0 openai/openai-node

by openai-sdks[bot]

The Node SDK's Live MCP support arrives as typed request parameters: server settings, allowed tools, and approval policies. Live MCP server configuration is a distinct addition from the budget removal, so read the release notes for which of your pins move.

0.170.0 openai/codex

by github-actions[bot]

Codex added WebRTC negotiation for realtime voice calls in the TUI, automatically refreshed auth headers for OpenTelemetry HTTP exporters, and experimental support for identically named MCP servers across environments. The exporter refresh matters if you ship Codex telemetry to a collector that enforces short-lived tokens.

Support remote app servers in `codex exec` openai/codex

by copyberry

codex exec --remote <ADDR> now runs a noninteractive turn against an existing app server's configuration and credentials, starting a new thread or resuming one by UUID, with human-readable or JSON output. It puts scripted runs on the same server the interactive client uses instead of a separate local setup.

Mask system SSH config drop-ins in the Linux sandbox (#61947) openai/codex

by viyatb-oai

The Linux sandbox masks /etc/ssh/ssh_config.d with an empty read-only mount. The reason given is that root-owned drop-ins can appear to belong to nobody inside the user namespace, which leads OpenSSH to reject them and SSH connections to fail.

Remove the 4,000-byte prompt limit for `thread/sample` (#62378) openai/codex

by Keyan Zhang

Requests longer than 4,000 bytes now pass thread/sample validation; expectedTurnId and prompt must still be nonblank. Long prompts reaching sampling no longer fail on length alone.

feat(api): support Live WebRTC standby openai/openai-python

by markstuart-oai

Live WebRTC sessions can be configured to pause when idle and resume on speech, via a speech wake policy, VAD threshold, and idle timeout at session creation, with session.paused and session.resumed delivered as typed sideband events. Requests can also read optional active and paused billing fields.

Include enclosing tool context in nested Guardian approvals openai/codex

by copyberry

Alongside: nested Guardian approvals now carry their enclosing tool context, bounded guardian retries landed, Codex traces moved to typed IDs, stale exec signatures and a docked Command Center closed out the week, and the Windows MXC sandbox moved to the public V1 SDK.

$ ls openai/week/ # the briefings behind this review

Keep up with OpenAI in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.

Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.

all openai reviews →