FOSSWire
Search/
Local AI & Edge

LocalAI 4.12.0 introduces redesigned UI and streaming voice responses

Brief written by the FOSSWire AI editor from LocalAI How we write

1 min read

Key Takeaway

LocalAI 4.12.0 adds a redesigned web UI, streaming voice capabilities, native decision routing, and persistent Qdrant vector stores.

Summary

The LocalAI 4.12.0 release brings significant updates to the self-hosted AI platform, including a complete web UI redesign with light and dark themes. New features include streaming voice responses via Qwen TTS, native decision routing for request classification, and image upscaling within the Studio workflow. The update also introduces persistent Qdrant stores and enhanced privacy controls like PII filtering.

Technical improvements in LocalAI 4.12.0

  • Audio & Voice: Supports incremental Qwen TTS, Parakeet bundles, and configurable Voice Activity Detection (VAD).
  • Decision Routing: Implements a /v1/decisions adapter to handle predicates, choices, and scores via decision models.
  • Inference Controls: Adds moe_cache_mib, lazy_mode, and explicit slot prompt-cache save/restore capabilities.
  • Storage & Vector DB: Integrates persistent Qdrant stores with TLS and environment-backed credentials via the /stores/* API.
  • Motion Capture: Includes a GEM-X API to convert RGB frames into SOMA-77 or SMPL-24 poses using CPU or Vulkan.

Why it matters

This release prepares LocalAI for a modular future by introducing addon support architecture, allowing components like Studio and Talk to be installed on-demand.

Original reference: github.com →