v4.0.0

Mudler has launched LocalAI 4.0, fundamentally transforming the open-source tool from a simple model runner into a comprehensive AI orchestration platform. The most significant addition is native agentic capabilities embedded directly into the core, allowing users to create, manage, and deploy AI agents with memory systems and skill databases. This is complemented by the launch of Agenthub, a centralized community space where users can share and import pre-built agents directly into their LocalAI instances. The platform now supports full agent lifecycle management through a completely revamped React-based interface, offering improved performance and modern UX.

The technical upgrades are substantial. LocalAI 4.0 introduces Canvas Mode, which lets users preview code blocks and artifacts side-by-side in the chat interface. It provides full client-side support for the Model Context Protocol (MCP), enabling tool streaming and allowing users to select which MCP servers to enable directly from the UI. New experimental backends include MLX Distributed for running distributed workloads on Apple silicon, along with audio backends like fish-speech and faster-qwen3-tts. The release also adds WebRTC support for low-latency real-time audio conversations and improves infrastructure with Podman documentation and persistent data path separation.