v0.5.13 @magnitudedev/cli@0.2.0
vadimdemedes/inkSep 30, 2026 → Sep 30, 20260 days between releases
What's New
Replaced llama.cpp engine with Magnitude's engine and introduced new server commands
Breaking Changes
- Replaced `magnitude service` commands with `magnitude serve` and `magnitude status`.
- Model and catalog commands now require an existing Desktop or server instance.
- Remote caller API key checks are now gated in route handlers.
New Features
- New Magnitude inference engine with efficient kernels.
- Server-only mode for headless inference.
- Improved speculative decoding speed and draft acceptance.
Improvements
- Speculative decoding significantly faster on Apple Silicon and NVIDIA GPUs.
- Automatic hardware optimization for inference.
Bug Fixes
- Fixed long prompt crashes on Apple Silicon.
- Fixed Gemma 4 and Qwen3.6 models loading on Apple Silicon.
- Fixed speculative drafts with sliding-window layers failing to load.
Upgrade Notes
Update scripts to use `magnitude serve` and `magnitude status` instead of `magnitude service`. Ensure a Desktop or server instance exists before running model or catalog commands. Review route handler logic for remote caller API key checks.
The Two Releases
Release v0.5.13
Sep 30, 2026· 22 days ago
A comprehensive patch release for opentui featuring numerous native buffer optimizations, including improved text encoding, buffer resizing, and terminal capability handling.
Full release notes@magnitudedev/cli@0.2.0
Sep 30, 2026· 22 days ago
A major version update replacing the llama.cpp engine with Magnitude's own optimized engine, introducing a new server command structure, and significantly improving speculative decoding performance.
Full release notes