●Change8

v3.2.0

Breaking Changes
📦 localaiView on GitHub →
⚠ 2 breaking✨ 8 features🐛 1 fixes🔧 5 symbols

Summary

LocalAI 3.2.0 introduces a major architectural shift by separating inference backends from the core binary, resulting in a leaner application and enabling independent backend management. This release also adds automatic hardware detection for backend installation and expands model support significantly.

⚠️ Breaking Changes

  • The core is now separated from all inference backends (llama.cpp, whisper.cpp, piper, stablediffusion-ggml, etc.). These are no longer bundled in the main binary.
  • Existing models installed prior to v3.2.0 might not have a specific backend assigned and may require manual backend installation after upgrading for them to function.

Migration Steps

  1. If you have existing models installed before upgrading to 3.2.0, you may need to install the required backend manually for these models to work. Use the WebUI or the CLI command: `local-ai backends install <backend_name>`.
  2. For advanced use cases or to override auto-detection of hardware capabilities, use the LOCALAI_FORCE_META_BACKEND_CAPABILITY environment variable (options: default, nvidia, amd, intel).

✨ New Features

  • Introduction of Modular Backends: Backends now live outside the main binary in a Backend Gallery, allowing independent updates.
  • Drastically smaller LocalAI binary and container images due to backend separation.
  • Smart Backend Installation: LocalAI automatically detects hardware (CPU, NVIDIA, AMD, Intel) and downloads the necessary backend upon model installation.
  • Simplified Build Process due to the new modular architecture.
  • Intel GPU Support for Whisper transcription via SYCL acceleration.
  • Enhanced Realtime Audio support with speech started and stopped events.
  • OpenAI-compatible support for the input_audio field in the /v1/chat/completions API for multimodal audio inputs.
  • Massive Model Expansion: Over 50 new models added, including Qwen3, Gemma, Mistral, and Nemotron.

🐛 Bug Fixes

  • Fixed an issue where the download status for backend images in the gallery was not correctly displayed.

🔧 Affected Symbols

llama.cppwhisper.cpppiperstablediffusion-ggml/v1/chat/completions