tool-pixi

What problem does it solve?

It solves the dependency-hell problem in ML projects by helping you create deterministic, hardware-aware Python environments that are consistent across machines and teams.

Core Features & Use Cases

• Deterministic environment replication: Use pixi.lock to lock exact package versions for repeatable experiments and deployments.
• GPU/CUDA compatibility checks: Define minimum system requirements (like CUDA and glibc) and let Pixi select compatible binaries before runtime failures occur.
• Unified conda + PyPI workflow: Mix conda-based system libraries with PyPI packages (often via uv) within one manifest for smoother ML setup.
• Multi-environment support: Define CPU/GPU and dev/test/prod configurations in one pixi.toml and install or run the exact environment you need.

Example use case: You want a training environment that matches a specific CUDA version for GPU runs, while keeping a separate CPU/MPS-friendly setup for local development and CI—using one manifest and locked outputs.

Quick Start

Use the tool-pixi skill to set up a new ML project environment by initializing pixi in your repository, adding both conda and PyPI dependencies, and installing to generate a reproducible pixi.lock file.