v3.8.30 v3.8.31
cheaper-inference/OmniRouteJun 20, 2026 → Jun 20, 20260 days between releases
What's New
Dashboard performance tuning and NVIDIA NIM embedding fixes
Breaking Changes
- Codex hooks flags are deprecated.
New Features
- One-click Redis & Bifrost sidecar provisioning.
Improvements
- Dashboard combos UI split into leaf components for performance.
- Next.js config tuning for standalone builds.
- Security hardening for OAuth callbacks.
- Refactoring of chatCore to extract heap-pressure guard.
Bug Fixes
- NVIDIA NIM embeddings now inject the required input_type.
- Fix for SSE response leaking null data events.
- Fix for stripping assistant echo fields on OpenAI targets.
- Fix for AI SDK-style image content parts acceptance.
Upgrade Notes
Codex hooks flags are deprecated; update your configuration accordingly. NVIDIA NIM embedding models now automatically inject the required input_type parameter.
The Two Releases
v3.8.30
Jun 20, 2026· 4 months ago
A major release introducing memory as OFF by default, a suite of new CLI setup commands for coding tools, and three new OpenAI-compatible providers (OpenAdapter, dit.ai, TokenRouter). It includes extensive bug fixes for provider model discovery, SSE streaming, and security hardening for DuckDuckGo scraping.
Full release notesv3.8.31
Jun 20, 2026· 4 months ago
This release focuses on dashboard performance improvements, specifically splitting the combos UI into leaf components. It includes critical fixes for NVIDIA NIM embeddings requiring `input_type`, deprecation of Codex hooks flags, and security hardening to scope OAuth callbacks to a trusted origin list. Maintenance updates include comprehensive documentation catch-up and refactoring of chatCore.
Full release notes