OpenAI Codex `rust-v0.124.0` adds Bedrock support and stable hooks
OpenAI Codex `rust-v0.124.0` ships Amazon Bedrock support with AWS SigV4 auth, stable hooks configurable via `config.toml`, TUI quick-reasoning keyboard shortcuts, and multi-environment app-server sessions, alongside a range of bug fixes.
Score breakdown
Teams using Codex with AWS infrastructure can now authenticate directly via Bedrock with SigV4, while stable hooks and multi-environment app-server sessions unlock more sophisticated agentic workflows without manual workarounds.
- 01First-class Amazon Bedrock support added, including AWS SigV4 signing and AWS credential-based auth.
- 02TUI quick-reasoning shortcuts: `Alt+,` lowers reasoning, `Alt+.` raises it; model upgrades reset reasoning to the new model's default.
- 03Hooks are now stable and configurable inline in `config.toml` and `requirements.toml`, with visibility into MCP tools, `apply_patch`, and long-running Bash sessions.
OpenAI Codex `rust-v0.124.0` delivers several notable new capabilities. Amazon Bedrock is now a first-class supported provider for OpenAI-compatible model access, with AWS SigV4 request signing and AWS credential-based authentication. The TUI adds quick reasoning-level controls — `Alt+,` lowers reasoning effort and `Alt+.` raises it — and when a user accepts a model upgrade, reasoning now resets to that model's default rather than carrying over stale settings. Hooks graduate to stable status in this release: they can be configured inline in `config.toml` and managed via `requirements.toml`, and they can now observe MCP tools, `apply_patch` edits, and long-running Bash sessions. App-server sessions gain support for multiple managed environments with per-turn environment and working directory selection, making multi-workspace and remote targeting more precise. Eligible ChatGPT plans also now default to the Fast service tier unless explicitly opted out.
On the bug-fix side, Cloudflare cookies are now preserved across approved ChatGPT hosts to reduce auth breakage in HTTP-backed flows.
On the bug-fix side, Cloudflare cookies are now preserved across approved ChatGPT hosts to reduce auth breakage in HTTP-backed flows. Remote app-server reliability is improved, with websocket events continuing to drain under load and shutdown no longer failing when the remote worker exits during cleanup. Permission-mode drift is fixed so `/permissions` changes survive side conversations and Full Access state is correctly reflected in MCP approval handling. `wait_agent` now returns promptly when mailbox work is already queued rather than waiting for a fresh notification or timing out. Local stdio MCP launches for relative commands without an explicit `cwd` now fall back to path resolution consistent with CLI behavior, and startup is more resilient to managed config edge cases, with unknown feature requirements warning instead of aborting.
Key facts
- 01First-class Amazon Bedrock support added, including AWS SigV4 signing and AWS credential-based auth.
- 02TUI quick-reasoning shortcuts: `Alt+,` lowers reasoning, `Alt+.` raises it; model upgrades reset reasoning to the new model's default.
- 03Hooks are now stable and configurable inline in `config.toml` and `requirements.toml`, with visibility into MCP tools, `apply_patch`, and long-running Bash sessions.
- 04App-server sessions now support multiple managed environments with per-turn environment and working directory selection.
- 05Eligible ChatGPT plans now default to the Fast service tier unless explicitly opted out.