v0.1.10 @magnitudedev/cli@0.2.2
facebook/astryxSep 30, 2026 → Oct 1, 20260 days between releases
What's New
@magnitudedev/cli v0.2.2 fixes stability issues and introduces API requirement changes compared to facebook/astryx v0.1.10
Breaking Changes
- count_tokens now requires a max_tokens argument
- Tool call IDs no longer repeat across turns
- Free-form object parameters now fail on Gemma 4
New Features
- Admission control mechanism
- Speculative decoding support
- Tool call ID generation fix
Improvements
- Model speed estimation uses memory bandwidth instead of GPU kernels
- System prompts are cached to avoid full re-reads
- Repeated images are encoded once to prevent errors
Bug Fixes
- Local models no longer hang under concurrent load
- Speculative decoding models no longer fail mid-request
- App no longer gets stuck on Assessing models on some hardware
- Anthropic token usage is now counted correctly
- Gemma 4 tool parameters are handled correctly
Upgrade Notes
Update code to provide max_tokens when calling count_tokens. Verify tool call handling logic as IDs are no longer repeated. Check parameter handling for free-form objects on Gemma 4.
The Two Releases
v0.1.10
Sep 30, 2026· 21 days ago
This update refreshes the application's UI with improved dark mode and adds city search volume to Keyword Research, along with support for scheduled rank checks and Shopify site audits.
Full release notes@magnitudedev/cli@0.2.2
Oct 1, 2026· 21 days ago
Fixes for local model hanging, speculative decoding failures, and various tool call issues to improve stability and performance.
Full release notes