v3.12.0
The open-source AI platform now handles text, images, and audio simultaneously with 40% faster processing.
Mudler's LocalAI platform has launched version 3.12.0, marking a significant upgrade for the open-source AI infrastructure tool. The release introduces multi-modal realtime capabilities, allowing users to send text, images, and audio simultaneously in conversations for richer interactions. A new Voxtral backend provides high-quality text-to-speech functionality, while enhanced multi-GPU support improves Diffusers performance for image generation tasks. The update also includes legacy CPU optimizations for older processors, improved UI themes with dark/light variants, and multiple stability fixes for audio, image, and model handling. LocalAI now integrates with the broader 'Local Stack' ecosystem including LocalAGI for agent orchestration, LocalRecall for knowledge bases, and new tools like Cogito (Go library for agentic software) and Wiz (terminal-based AI assistant). The release addresses numerous bug fixes including security validation for URLs, websocket locking issues, and excessive logging problems.
- Multi-modal realtime conversations supporting simultaneous text, image, and audio inputs for richer AI interactions
- New Voxtral text-to-speech backend and improved multi-GPU support for 40% faster Diffusers performance
- Enhanced legacy CPU compatibility and UI themes plus integration with Local Stack ecosystem tools
Why It Matters
Enables privacy-first, locally-run AI with enterprise-grade multimodal capabilities previously only available in cloud services.