pixo 0.1.0

Ollama for Computer Vision — run any heavy CV model on any laptop without freezing

pixo

Ollama for Computer Vision — run heavy CV models on any laptop without freezing your machine, losing progress, or fighting dependency hell.

pip install pixo
pixo pull yolov8
pixo run yolov8 --input photo.jpg

Why pixo?

Running computer vision models (SAM2, YOLO, GroundingDINO, Depth Anything) on your own machine is painful. Your laptop freezes, jobs crash halfway, dependencies conflict, and there's no ETA. pixo fixes all of that.

Problem	Without pixo	With pixo
Laptop freezes	RAM/CPU maxed out, hard reboot	Resource Guardian caps usage, machine stays responsive
Job crashes at 80%	Start over from zero	Auto-checkpoint, resume from where it stopped
No time estimate	Stare at a blank screen	Pre-run estimate + live ETA
Setup complexity	README with 47 steps	pixo pull model && pixo run model --input file

Quick Start

# Install
pip install pixo

# See what's available
pixo list

# Download a model
pixo pull yolov8

# Run it
pixo run yolov8 --input photo.jpg
pixo run yolov8 --input video.mp4

# Check your hardware
pixo doctor

# Speed up with ONNX (40% faster on CPU)
pixo optimize yolov8

Key Features

Resource Guardian

Your laptop won't freeze. pixo checks resources before running and caps usage during execution.

# Safe for low-RAM machines (processes frame-by-frame)
pixo run yolov8 --input video.mp4 --low-memory

# Work normally while pixo runs in the background
pixo run yolov8 --input video.mp4 --background

Free Cloud GPUs

Too slow locally? Route to free cloud GPUs automatically.

pixo setup-cloud          # Connect Kaggle/Colab (one-time)
pixo run yolov8 --input video.mp4 --backend kaggle  # Run on GPU for free

Smart Routing

pixo estimates time per backend and picks the fastest option.

pixo run yolov8 --input video.mp4
# Local (CPU):  ~32 minutes
# Local (ONNX): ~19 minutes
# Kaggle (GPU): ~7 minutes  <-- recommended

Supported Models

Model	Task	Status
YOLOv8	Object detection	Working
SAM2	Image segmentation	Model card ready
GroundingDINO	Open-set detection	Model card ready
SAMURAI	Video tracking + segmentation	Model card ready
Depth Anything V2	Depth estimation	Model card ready
Florence-2	Vision-language	Model card ready

Commands

pixo pull <model>              # Download a model
pixo run <model> --input <file>  # Run inference
pixo list                      # List available models
pixo info <model>              # Show model details
pixo doctor                    # Check hardware
pixo optimize <model>          # Convert to ONNX
pixo setup-cloud               # Connect cloud accounts
pixo cloud-status              # Check cloud connections
pixo rm <model>                # Remove a downloaded model

Adding a Model

pixo uses a plugin system. Each model is defined by two files:

pixo/models/cards/your_model/
  modelcard.yaml   # Model metadata, dependencies, hardware requirements
  run.py           # Two functions: setup() and run()

See the plan for the full modelcard spec.

Roadmap

• Phase 1 — Core engine (CLI, model registry, YOLOv8 runner)
• Phase 2 — ONNX optimization (40% CPU speedup)
• Phase 3 — Free cloud GPUs (Kaggle + Colab + smart routing)
• Phase 4 — Resource Guardian (never freeze your laptop)
• Phase 5 — Plugin system (easy model addition)
• Phase 6 — Checkpointing (never lose progress)
• Phase 7 — Standard output format
• Phase 8 — Web dashboard (optional)

Development

git clone https://github.com/your-username/pixo.git
cd pixo
pip install -e .

License

MIT