1minds3t/omnipkg repository overview

⁠omnipkg — Python Runtime Hypervisor

Break the fundamental laws of Python environments. Run infinite Python versions, package combinations, and PyTorch ABIs simultaneously in one container. Built for high-performance ML infrastructure and CI/CD.

⁠🚀 Quick Start

# Latest stable (Debian-based, most compatible)
docker pull 1minds3t/omnipkg:latest

# Run interactive shell
docker run -it 1minds3t/omnipkg:latest

# Try the famous Daemon IPC Showcase (True Zero-Copy GPU)
docker run -it 1minds3t/omnipkg:latest omnipkg demo 11

⁠🎯 Why omnipkg in Docker?

The "One Image" CI/CD Matrix: Stop building and caching separate 2GB Docker images for python3.9, python3.11, and python3.12. Pull one omnipkg image, and let the C-dispatcher hot-swap the Python interpreter and dependency trees in <180ms.

Production-Grade Performance:

• ⚡ 237µs Startup Time: A custom C-dispatcher handles command routing, executing 17x faster than uv and 3x faster than bun before Python even wakes up.
• 🔥 Universal CUDA IPC ("God Mode"): Pure ctypes VRAM-to-VRAM GPU transfers. Pass a 4MB tensor through 3 PyTorch versions in 4.5ms (1,107× faster than standard multiprocessing).
• ⚛️ Atomic Stash-Swaps: Rust FFI intercepts uv install plans to perform atomic os.rename() directory swaps in <1ms.

Multi-Version Concurrency:

• Run Python 3.7–3.15 simultaneously in one container.
• Install completely conflicting ABIs (torch==1.13.1+cu116 and torch==2.2.0+cu121) in the same environment, and execute them concurrently without data leaving VRAM.
• Auto-healing: Detects & installs missing dependencies 7-12× faster than UV.

⁠📊 Benchmarks (Real Production Workloads)

Benchmarks validated live via omnipkg stress-test and omnipkg demo 11.

⁠📦 Available Docker Tags

⁠Version-Specific (Current)

• 3.4.x – Latest release versions
• 3.4.x-debian, 3.4.x-alpine – Version + base OS

⁠Production

• latest, debian – Most compatible (glibc)
• slim, alpine – Minimal footprint, musl libc (~50MB)
• ubuntu-20.04, ubuntu-22.04, ubuntu-24.04
• fedora – RHEL/Fedora-based

⁠Specialized

• cuda, gpu – NVIDIA GPU support (Universal CUDA IPC + Torch MP Queue fallback built-in)
• dev – Development tools included

⁠🏗️ Multi-Architecture Support

✅ linux/amd64 (x86_64)
✅ linux/arm64 (aarch64)

Docker automatically pulls the correct architecture for your platform.

⁠💡 Common Use Cases

⁠CI/CD Testing (Eliminate the Matrix)

Instead of a giant GitHub Actions matrix spawning 5 containers, run one:

docker run --rm 1minds3t/omnipkg:latest bash -c "
  8pkg39 run pytest && \
  8pkg311 run pytest && \
  8pkg313 run pytest"

⁠GPU-Accelerated ML Workloads

docker run --gpus all -it 1minds3t/omnipkg:cuda bash

⁠Production Scripts with Auto-Healing

If an upstream deployment breaks a dependency, omnipkg run detects the ImportError or ABI crash, finds the correct bubbled version, and executes it anyway.

docker run --rm -v $(pwd):/app 1minds3t/omnipkg:alpine 8pkg run /app/script.py

⁠🎓 Real-World Example: Daemon IPC

from omnipkg.isolation.worker_daemon import DaemonClient

client = DaemonClient()

# Execute code in PyTorch 2.0.1 (Worker 1)
client.execute_smart("torch==2.0.1+cu118", """
import torch
print(f"Running on {torch.cuda.get_device_name(0)} with Torch {torch.__version__}")
""")

# Instantly switch to PyTorch 2.2.0 (Worker 2, shared VRAM)
client.execute_smart("torch==2.2.0+cu121", "import torch; print(torch.__version__)")

⁠🔧 Environment Variables

• OMNIPKG_HOME=/home/omnipkg – User home directory
• OMNIPKG_NONINTERACTIVE=1 – Disables all prompts for CI/CD environments
• OMNIPKG_LANG=es – Set CLI language (supports 24+ languages)

⁠🔗 Links

• GitHub – github.com/1minds3t/omnipkg⁠
• PyPI – pypi.org/project/omnipkg⁠
• Conda-Forge – anaconda.org/conda-forge/omnipkg⁠
• Interactive Docs – 1minds3t.echo-universe.ts.net/omnipkg⁠

⁠🤝 Contributing & License

Bugs or feature requests? Open an issue on GitHub or contact: [email protected]⁠

Dual-licensed: AGPLv3 (open source) / Commercial (enterprise)

Stop fighting dependency hell. Start shipping production ML infrastructure.