Pyaket – Easy Python to → Fast Executables

4 months ago 4

📦 Pyaket is a tool that bundles and generates portable executables of your python projects for all platforms. No more convoluted installation steps, give users the convenience they want, with maximum compatibility and dev-centric ease of use compared to alternative solutions.

Lightning fast installation that automatically manages python, virtual environments, and dependencies without user intervention that just works, bundle wheels or install from pypi.
Max compatibility with how the project is run in the user's machine - pyaket does not reinvent the wheel or compile python with an intermediate, use tools that already exists #
Cross compile from anywhere to most platforms and architectures, no docker or virtual machines required, portable immutable executables - see the table for details! #
Intelligently detects partial installations, downloads, archive unpacks, and automatically takes appropriate action - making iterative development easy and resilient against users
Standalone executables with no network calls at runtime that bundles all dependencies #
Monorepo support in mind, decoupled dependencies and entry point specification
Rolling releases where a single binary always runs latest pypi or git branch/tag #
PyTorch installation at runtime, automatic backend detection (optional). #

Compile a cowsay binary for the current platform and run it:

$ pyaket app --name cowsay --pypi "cowsay==6.1" run --module cowsay compile Compiling libc v0.2.172 Compiling typenum v1.18.0 ... Finished `release` profile [optimized] target(s) in 9.88s

$ ./release/cowsay-linux-amd64-v0.0.0.bin -t "Hello, Pyaket!" ______________ | Hello, Pyaket! | ============== \ \ ^__^ (oo)\_______ (__)\ )\/\ ||----w | || ||

Fast iterative development

with a warm build cache:

$ pyaket app -n cowsay -p "cowsay==6.1" run -m cowsay compile Finished `release` profile [optimized] target(s) in 1.54s

after the first installation:

$ hyperfine "./release/cowsay-linux-amd64-v0.0.0.bin -t anyhow" Time (mean ± σ): 23.3 ms ± 0.3 ms [User: 15.8 ms, System: 7.2 ms] Range (min … max): 22.9 ms … 24.8 ms 100 runs $ hyperfine "python -m cowsay -t anyhow" Time (mean ± σ): 18.5 ms ± 0.1 ms [User: 14.2 ms, System: 4.1 ms] Range (min … max): 18.2 ms … 19.0 ms 100 runs

^{Note: For the keen among you, the actual benchmark command was nice -20 taskset -c 2 hyperfine -w 50 -r 100 -N (...), executed on Python 3.13.3, mainline Linux kernel v6.14.4, R9 5900x ondemand governor stock, 2x3200 MT/s DDR4 CL16 2Rx8 as of May 2025}

to most platforms and architectures easily:

# Windows executables compiled from linux $ pyaket app -n cowsay -p "cowsay==6.1" run -m cowsay release -t windows compile Finished `release` profile [optimized] target(s) in 8.11s $ wine ./Release/cowsay-windows-amd64-v0.0.0.exe -t "Hello, Wine!" ____________ | Hello, Wine! | ============ \ \ ^__^ (oo)\_______ (__)\ )\/\ ||----w | || ||

# Intel Macbook @ ./release/cowsay-macos-amd64-v0.0.0.bin $ pyaket ... release --target macos --arch amd64 compile # Apple Silicon @ ./release/cowsay-macos-arm64-v0.0.0.bin $ pyaket ... release --target macos --arch arm64 compile

and install them at runtime, perfect for monorepos:

$ uv build --all-packages --wheel -o dist Successfully built dist/shared-1.0.0-py3-none-any.whl Successfully built dist/project_a-1.0.0-py3-none-any.whl Successfully built dist/project_b-1.0.0-py3-none-any.whl # Both will share the same virtual environment 🤯 # ./release/{project_a,project_b}-linux-amd64-v0.0.0.bin $ pyaket app -n project_a -w "dist/*.whl" run -m project_a compile $ pyaket app -n project_b -w "dist/*.whl" run -m project_b compile

at runtime, with automatic backend detection:

# ./release/app-linux-amd64-v0.0.0-auto.bin $ pyaket ... torch -v 2.7.0 -b auto compile # ./release/app-linux-amd64-v0.0.0-cu128.bin $ pyaket ... torch -v 2.7.0 -b cu128 compile

For more examples, proper configuration and advanced features, check out the website page!