AFM is a 100% Swift wrapper for MLX with advanced inference features for Apple Silicon. This automation-ready framework enables telegram integration for remote model access, experimental tool parsing with afm_adaptive_xml, prefix caching for KV cache reuse, grammar-constrained decoding for tool calls, concurrent batch processing, and guided JSON output. Perfect for building AI automation workflows that require reliable tool calling, batch inference, and remote access. Features include --enable-prefix-caching for performance, --enable-grammar-constraints for structured output, --concurrent mode for parallel requests, and --guided-json for schema compliance. No Python required - pure Swift implementation with pip installability.
