omlx
About
oMLX is a high-performance LLM inference server optimized for Apple Silicon, featuring a unique tiered KV cache system that persists context across RAM and SSD. Built on the MLX framework, it supports continuous batching, multi-model serving with LRU eviction, and native VLM/OCR capabilities. It acts as a critical backend for the OpenClaw ecosystem, offering OpenAI-compatible APIs and specific optimizations for Claude Code to enable efficient local agent workflows.
Features
- Tiered KV Cache (Hot/Cold)
- Continuous Batching Support
- Native Apple Silicon Optimization
- VLM & OCR Capabilities
- Deep OpenClaw Integration
Supported Platforms
Links
Related AI Industry News
Accenture and Google Cloud Launch Gemini Enterprise Acceleration Program
Accenture and Google Cloud have unveiled the Gemini Enterprise Acceleration Program to help organizations scale AI from prototypes to core operations. The initiative features dedicated Forward Deployed Engineers and exclusive access to Google DeepMind’s latest models.
Read Full Story →
Cline Launches Open-Source Agent Runtime SDK, Outperforming Anthropic
Cline has open-sourced its internal Agent runtime as @cline/sdk, a modular four-layer stack supporting plugins, subagents, and MCP. It notably outperformed Anthropic's Claude Code on the Terminal Benchmark 2.0.
Read Full Story →
Claude Code v2.1.141 Adds Desktop Notifications and Context Summarization
Anthropic has released Claude Code v2.1.141, introducing terminalSequence hooks for reliable desktop notifications and a context summarization feature. These updates enhance productivity by allowing developers to track background tasks and manage long context windows efficiently.
Read Full Story →