Local Multimodal LLM v0.1.0-beta

Local Multimodal LLM for Unreal Engine 5.8

First public beta from Ian Sundahl / Volley Studios. This release provides private, offline character dialogue with a native llama.cpp LLM runtime, Parakeet speech recognition, Pocket TTS, multi-character sessions, relationship context, dialogue routing, guarded streaming, and constrained Unreal tool calls.

Recommended download

• Starter Core (Win64): recommended default; includes CPU and Vulkan backends plus Gemma 4 E2B, Parakeet, Pocket TTS, and four starter voices.
• Starter NVIDIA (Win64): includes the same starter stack plus the CUDA backend and NVIDIA runtime dependencies.
• Starter Model Pack: models and voices only, for users who already have the plug-in runtime.

Download every .partNNN file for one package, Reassemble-And-Extract.ps1, and its matching Extract-Starter-*.cmd file into the same directory. Double-click the matching extraction command. Use RELEASE_ASSET_SHA256SUMS.txt to verify downloads.

Status and limitations

• Beta release targeting Unreal Engine 5.8 on Win64.
• Core was compiled and validated through an isolated clean-project Shipping workflow on the development machine.
• CUDA, Vulkan, and CPU backends are bundled, but separate physical-machine certification remains outstanding.
• Vision/mmproj integration is available as a development feature, not part of the starter path.
• Chatterbox and NeuTTS adapters remain optional/experimental and are not included in the starter model pack.

See the included README, documentation, licenses, notices, and attribution files before redistribution.