v3.7.1 v3.7.2
cheaper-inference/OmniRouteApr 27, 2026 → Apr 28, 20261 day between releases
What's New
Adds centralized authz pipeline and native Codex websockets for beta models
Breaking Changes
- Clients may receive native websockets instead of HTTP responses for beta-gated Codex models.
- Requests exceeding 128 tools will now fail with 400 errors in executors.
- Centralized authz pipeline may require configuration updates for custom auth logic.
New Features
- Centralized proxy-based authorization pipeline and lifecycle policy.
- Configurable call log pipeline artifacts.
- Native Codex websocket responses on beta-gated models.
- Guarded remote image fetch utility.
- Continue meta.ai conversations across turns.
Improvements
- Replace unsupported Bottleneck maxWait option.
- Add body-read timeout to prevent pending request hangs.
- Guarded remote image fetch utility for security.
Bug Fixes
- Sanitize empty string placeholders in tool-call arguments.
- Prevent protocol leakage and fabricated instructions on bare chat requests.
- Truncate tools array to 128 items max in executors.
- Add body-read timeout.
- Replace unsupported Bottleneck maxWait option.
Upgrade Notes
Clients using beta-gated Codex models will now receive native websocket responses. Ensure the tools array passed to executors does not exceed 128 items to avoid 400 errors. The centralized authz pipeline may require updated configuration if custom auth logic was previously implemented.
The Two Releases
v3.7.1
Apr 27, 2026· 6 months ago
A maintenance release focusing on GPT-5.5 support, extensive i18n expansion, and critical bug fixes including a CLI recovery command for encryption issues.
Full release notesv3.7.2
Apr 28, 2026· 6 months ago
Introduces a centralized proxy-based authorization pipeline, call log pipeline configuration, and enables native Codex websocket responses.
Full release notes