The Universal AI Runtime
Run any model. Any modality. Any hardware. One engine.
Website | GitHub | Model Hub | PyPI
One runtime that executes any AI model — image generation, language models, audio, video, vision-language — through a single CLI, a single container format (.nbx), and automatic hardware allocation.
pip install neurobrix
neurobrix import sana/1600m-1024 --no-keep
neurobrix run --model Sana_1600M_1024px_MultiLing --prompt "A sunset over mountains"Runs on Linux, Windows, and macOS. Auto-detects your GPU.
| Ollama | ComfyUI | vLLM | NeuroBrix | |
|---|---|---|---|---|
| LLMs | Yes | -- | Yes | Yes |
| Image Generation | -- | Yes | -- | Yes |
| Video Generation | -- | -- | -- | Yes |
| Audio (STT + TTS) | -- | -- | -- | Yes |
| Multi-GPU Auto-Allocation | -- | -- | Yes | Yes |
| Universal Model Format | -- | -- | -- | NBX (any) |
| Cross-Platform | Yes | -- | -- | Yes |
Video (11): Wan2.1 T2V and VACE, CogVideoX, Allegro, Open-Sora, SANA-Video, Mochi Upscaling (8): Real-ESRGAN, HAT, SwinIR, Swin2SR Image (5): Sana (1K/4K), PixArt-Alpha, PixArt-Sigma, Flex.1-alpha Multimodal (5): Qwen3-Omni, Ming-Lite-Omni, Janus-Pro Speech synthesis (5): Kokoro, VibeVoice, Orpheus, OpenAudio S1, Chatterbox Speech recognition (3): Whisper Large V2, Whisper V3 Turbo, Parakeet Audio language (3): Canary-Qwen, Granite Speech, Voxtral Language (3): DeepSeek-MoE-16B, Qwen3-30B-A3B, TinyLlama Code (3): DeepSeek-Coder-V2-Lite, Qwen3-Coder-30B-A3B Vision-language (1): Qwen3-VL-30B-A3B
Browse: neurobrix.es/models
| Repository | Description | Status |
|---|---|---|
| neurobrix | Core runtime, CLI, Prism solver, NBX format | Public Alpha |
| neurobrix-studio | Desktop application for macOS, Windows and Linux — a client of the engine, never a second way to run models | In development |
Apache 2.0 — contributions welcome. See the Contributing Guide.
© 2025-2026 · Developed by Hocine Benkelaya · Managed by WizWorks OÜ, a property of NEURAL NETWORK HOLDING LTD