@magnitudedev/cli@0.2.4 v0.3.0
cheaper-inference/OmniRouteOct 2, 2026 → Oct 2, 20260 days between releases
What's New
Introduces upstream Laya runtime fixes, new confidence fields, and improved truncation logic.
New Features
- Support for custom noul labels and boolean criteria keys
- Addition of `answer_confidence` alongside existing `confidence` semantics
- Enhanced language detection and email cleaning for multilingual routing
- Incremental preload improvements and prompt-prefix LRU protection
- Improved named input-validation error reporting
Improvements
- Conversation handling now preserves the newest turns under truncation
- Unicode is preserved in structured instructions
- State truncation and collapsed option spans are now reported
- Handles empty or language-neutral hints correctly during preload
- Protects prompt-prefix LRU under concurrent preparation
Bug Fixes
- Fixes zero-room left-truncation edge case in conversation handling
- Fixes handling of empty or language-neutral hints during preload
- Fixes state truncation and collapsed option spans reporting
Upgrade Notes
Review the new `answer_confidence` field alongside existing `confidence` and be aware of changes to conversation truncation logic. Custom criteria keys are now strictly validated. Ensure your configuration handles the new truncation behavior correctly.
The Two Releases
@magnitudedev/cli@0.2.4
Oct 2, 2026· 20 days ago
@magnitudedev/cli v0.2.4 improves kernel tuning efficiency and fixes memory issues and error reporting for inference failures.
Full release noteslaya-mlx v0.3.0
Oct 2, 2026· 20 days ago
Selectively brings upstream Laya runtime fixes into the native MLX port, focusing on conversation handling, truncation logic, and validation improvements.
Full release notes