pepip 0.0.1

Shared global environment package installer using symlinks — like pnpm for Python

pepip

Alternative to "conda" to share each dependency version in a shared global environment to save storage space

Overview

pepip (performance pip) solves a common pain point in Python development: each project typically has its own virtual environment, so large packages like torch or transformers are downloaded and stored multiple times — once per project.

pepip takes inspiration from pnpm for Node.js:

• Packages are installed once into a single shared global virtual environment (~/.pepip/global-venv), using uv for fast downloads.
• Inside each project's .venv, symlinks point back to the global copies, so Python can import the packages as normal.
• Re-installing a package that is already in the global environment is near-instant — no download required.

Requirements

• Python 3.8+
• uv — installed automatically as a dependency

Installation

pip install pepip

Usage

Install packages

# Install one or more packages
pepip install numpy pandas

# Install from a requirements file
pepip install -r requirements.txt

# Use a custom local venv path (default: .venv)
pepip install numpy --venv /path/to/my-env

After running pepip install, a .venv directory is created (or updated) in the current directory. Activate it as usual:

source .venv/bin/activate
python -c "import numpy; print(numpy.__version__)"

How it works

~/.pepip/
└── global-venv/
    └── lib/
        └── python3.12/
            └── site-packages/
                ├── numpy/          ← real package files (downloaded once)
                ├── numpy-2.0.dist-info/
                ├── torch/
                └── ...

my-project/
└── .venv/
    └── lib/
        └── python3.12/
            └── site-packages/
                ├── numpy -> ~/.pepip/global-venv/lib/.../numpy   (symlink)
                ├── numpy-2.0.dist-info -> ...                   (symlink)
                └── ...

The global environment path can be overridden via the PEPIP_HOME environment variable:

PEPIP_HOME=/shared/team-env pepip install torch

Development

# Clone and install in editable mode
git clone https://github.com/perf-pip/pepip
cd pepip
pip install -e .

# Run tests
pip install pytest
pytest

Evaluation — pepip vs uv

The eval/benchmark.py script measures installation latency and disk usage when setting up N projects with the same packages, comparing pepip against a plain uv workflow.

# 3 projects, default packages (tomli, packaging)
python eval/benchmark.py

# 5 projects, larger dependency set
python eval/benchmark.py --projects 5 --packages requests certifi charset-normalizer idna urllib3

# Keep the temp directories for manual inspection
python eval/benchmark.py --no-cleanup

Sample results (3 projects, packages: tomli packaging)

┌──────────────┬─────────────────┬─────────────┬──────────────────────┐
│  pepip vs uv — evaluation (3 project(s), packages: tomli packaging)  │
├──────────────┬─────────────────┬─────────────┬──────────────────────┤
│  Metric      │  uv (baseline)  │  pepip       │  Improvement         │
├──────────────┼─────────────────┼─────────────┼──────────────────────┤
│  Latency     │  0.34 s         │  0.35 s     │  +3.5 %              │
│  Disk usage  │  2.98 MB        │  1.10 MB ★  │  -63.2 %             │
└──────────────┴─────────────────┴─────────────┴──────────────────────┘

  ★ = better result

Sample results (5 projects, packages: requests certifi charset-normalizer idna urllib3)

┌──────────────┬─────────────────┬─────────────┬───────────────────────────────────────────────────────┐
│  pepip vs uv — evaluation (5 project(s), packages: requests certifi charset-normalizer idna urllib3)  │
├──────────────┬─────────────────┬─────────────┬───────────────────────────────────────────────────────┤
│  Metric      │  uv (baseline)  │  pepip       │  Improvement                                          │
├──────────────┼─────────────────┼─────────────┼───────────────────────────────────────────────────────┤
│  Latency     │  0.76 s         │  0.50 s ★   │  -33.8 %                                              │
│  Disk usage  │  9.46 MB        │  2.06 MB ★  │  -78.2 %                                              │
└──────────────┴─────────────────┴─────────────┴───────────────────────────────────────────────────────┘

  ★ = better result

Key observations:

• Storage savings are consistent regardless of project count: because each package version is stored exactly once in the global venv, the local .venv directories contain only tiny symlinks (~dozens of bytes each) instead of full copies. With 5 projects and the requests dependency tree (~2 MB each), pepip saves 7.4 MB compared to separate venvs.

• Latency savings grow with project count: the first project pays the same download + install cost as plain uv. Every subsequent project only needs venv creation + symlink creation, which is nearly instant. For tiny packages (cached by uv), the difference is negligible at n=3, but reaches −34 % at n=5 with a larger dependency set. For real-world packages like torch or transformers (GB-scale), the savings per extra project would be proportionally larger.