March 2026 monthly summary for Blaizzy/mlx-vlm: Key accomplishments include delivering a tool-driven server completion framework, memory-management optimization via a new prefill-step-size option, and improved tool-call traceability with an OpenAI-compliant index. These changes enable safer, scalable external tool integration, reduce memory-related failures, and improve observability, positioning the project for higher runtime throughput and reliability. No major bugs fixed this period.

February 2026 (ml-explore/mlx-lm): Focused on delivering a robust Pythonic function call parser for LFM2 models and stabilizing tool invocation UX. Implemented a Pythonic Function Calls Parser to interpret and execute Python-style tool calls within LFM2 workflows, added unit tests to validate behavior, and fixed a 404 error affecting content with non-thinking (instruct) models. These changes improve command interpretation, reliability, and maintainability, laying groundwork for broader automation and safer model-driven actions.