Real-Time Hyper-Modal AI for XR/VR/Robotics

Ultra-low latency language models with spatial awareness for immersive experiences and robotics applications.

🌐 Live Site: zenlm.org
🤗 HuggingFace: huggingface.co/zenlm
💻 GitHub: github.com/zenlm

Zen LM delivers real-time, spatially-aware AI optimized for:

• XR/VR Applications: Immersive experiences with sub-10ms latency
• Robotics: Multimodal perception and control
• Edge Deployment: Efficient models from 0.6B to 32B parameters
• Spatial Computing: Native 3D understanding and spatial audio

24+ Open-Source Models spanning:

• zen-nano (0.6B) - Ultra-efficient edge deployment
• zen-eco (4B) - Balanced performance with instruct/agent/coder/thinking variants
• zen-omni (7B) - Multi-task versatility
• zen-pro (7B) - Professional-grade with instruct/agent/thinking variants
• zen-coder (14B) - Code generation specialist (128K context)
• zen-next (32B) - Frontier performance

• zen-designer - Vision-language understanding
• zen-artist - Text-to-image generation and editing
• zen-video - Text/image-to-video generation
• zen-3d - 3D asset generation
• zen-world - 3D environment generation
• zen-voyager - Image-to-video with camera control
• zen-musician - Music generation
• zen-foley - Video-to-sound effects
• zen-scribe - Speech-to-text (multilingual)
• zen-director - Unified image/video generation with fine control

• zen-agent - Autonomous agent framework with tool use
• zen-code - IDE integration (VS Code, IntelliJ)
• zen-guard - Content moderation and safety
• zen-embedding - Semantic search and retrieval

• zen-engine - High-performance Rust inference engine
• zen-gym - Training framework (SFT, DPO, RLHF)
• zen-family - Model hub with conversion tools
• zen-blog - Technical documentation site

This repository hosts the zenlm.org documentation site:

• Home (index.html) - Mission, overview, and XR/VR focus
• Models (models.html) - Complete model catalog
• Research (research.html) - Papers and technical reports

zen/
├── docs/                      # Website root
│   ├── index.html            # Landing page
│   ├── models.html           # Model catalog
│   ├── research.html         # Research & papers
│   ├── papers/               # Research PDFs (15 papers)
│   │   ├── zen-technical-report.pdf
│   │   ├── zen-eco.pdf
│   │   ├── zen-omni.pdf
│   │   └── ...
│   └── assets/
│       ├── css/style.css     # Styling
│       ├── js/main.js        # Interactivity
│       └── logo.png          # Zen logo
├── .github/workflows/
│   └── pages.yml             # GitHub Pages deployment
└── README.md                 # This file

✅ Sub-10ms Latency - Optimized for real-time interaction
✅ Spatial Awareness - Native 3D understanding and spatial audio
✅ Multimodal Fusion - Vision, audio, video, 3D, and text
✅ Open Source - Fully transparent and permissively licensed
✅ Multiple Formats - SafeTensors, GGUF, MLX for any platform
✅ Edge to Cloud - Deploy anywhere from embedded to datacenter

Visit zenlm.org/models.html to browse the complete catalog.

# Example: Download zen-nano
huggingface-cli download zenlm/zen-nano

All technical papers available at zenlm.org/research.html:

• Zen Technical Report
• Individual model papers (15 PDFs)
• Training methodologies
• Benchmark results

Research papers are written in LaTeX and compiled to PDF:

cd /path/to/model/paper
pdflatex paper.tex
pdflatex paper.tex  # Run twice for references

Compiled PDFs are copied to docs/papers/ for the website.

# Serve the site locally
cd docs
python -m http.server 8000
# Visit http://localhost:8000

The site auto-deploys via GitHub Actions on push to main:

• Workflow: .github/workflows/pages.yml
• Live URL: https://zenlm.org
• Custom domain configured via docs/CNAME

This Repo (zenlm/zen): Documentation website only

• Website source code
• Research papers (PDFs)
• Deployment workflows

Model Repos: Individual GitHub repos for each model

• Training code
• Model weights on HuggingFace
• Evaluation scripts
• Model-specific documentation

Parent Directory (~/work/zen/): Development workspace

• Model training scripts
• Datasets
• Quantization tools
• Build artifacts

We welcome contributions! Areas of focus:

• Performance: Latency optimizations for XR/VR
• Spatial AI: Enhanced 3D understanding
• Multimodal: Better cross-modal fusion
• Robotics: Real-world deployment examples

If you use Zen LM in your research, please cite:

@techreport{zen2025,
  title={Zen: Ultra-Efficient Language Models for Local Deployment and Privacy Preservation},
  author={Zen Authors},
  institution={Zen LM},
  year={2025},
  url={https://zenlm.org}
}

All models and code are released under permissive open-source licenses. See individual model repositories for specific license details.

• Website: zenlm.org
• GitHub: github.com/zenlm
• HuggingFace: huggingface.co/zenlm

© 2025 Zen Authors. Built with clarity and purpose.