Hey there, John here, CTO at Pruna AI. Today, I'm thrilled to share some big news: Pruna AI is now available as a free(mium) package! 🎉

This release is a major milestone for us, and it's actually the result of a 3-in-1 project that we've been quietly working on since late September. Let me break it down:

1. We developed an internal token-generation engine to make our product accessible to you directly.

2. We revamped our entire documentation stack to create a better user experience.

3. We refactored Pruna AI's codebase for this new distribution channel.

It was a two-month sprint, powered by 50 PRs. Huge kudos to Johanna for tackling the documentation revamp and improving the smashconfig along the way. As for me, I rolled up my sleeves to build the backend and rework our codebase for compatibility. Let’s dive into the story behind this milestone and what it means for you.

TL;DR

Why we opened the doors to our CLI

Here’s the honest truth: using Pruna was annoying. For the longest time, getting started with our platform meant:

• Reaching out to us directly.

• Signing an NDA.

• Waiting for an access key.

• Going through a guided installation with our team.

Not exactly what I’d call a developer-friendly experience! Plus, we realized something: we weren’t channeling enough interest in the product. People were landing on our website, intrigued, but had no easy way to actually try out Pruna. A “Contact Us” form? Not exactly inspiring.

Now, Pruna AI is still proprietary (for now), but we wanted to extend its accessibility. Maybe this is a stepping stone toward open-sourcing it—who knows? For now, though, the goal is clear: let anyone explore our optimization engine without friction.

Removing the sales wall: just a pip-install away

With this release, you don’t need us anymore (in a good way!). No gatekeeping. No NDAs. No waiting. Just pip install and get started in your own environment.

pip install pruna[gpu]==0.1.1 --extra-index-url <https://prunaai.pythonanywhere.com/>

We also overhauled our documentation (finally!) to make sure it’s up-to-date, visually clean, and packed with enough detail to keep both beginners and power users happy. One of our top priorities was ensuring users could get the most out of Pruna by having access to comprehensive information about all available optimization methods.

Now, the documentation includes everything you need to know—like prerequisites (e.g., hyperparameters, CPU compatibility)—for two main categories:

• Open-source methods such as ctranslate, hqq , or torch.compile.

• Proprietary methods—like our x-fast compilation method, a general GPU compilation method that works on any model.

We believe in giving credit where it’s due, so every method in the docs clearly states its contributors (researchers and developers alike).

Cherry on the cake, it includes Colab notebooks (see “3x Faster Stable Diffusion Models” example) so you can test everything hands-on.

Now, the Pruna Package is accessible with just a token

On the backend, we’ve implemented a token system. The Pruna token serves as your unique identifier when using the platform. Upon your first call to the smash function, your token is automatically generated and printed in the console—no extra steps needed.

import torchvision
from pruna import smash, SmashConfig

# load any model
model = torchvision.models.vit_b_16(weights="ViT_B_16_Weights.DEFAULT").cuda()
smash_config = SmashConfig() # any SmashConfig, even an empty one, will do

# calling smash without a token will generate a new one
smashed_model = smash(model=model, token=None, smash_config=smash_config)

# ... the token will be printed in the console

# from now on, you can call smash with your token
smashed_model = smash(model=model, token="<your_token>", smash_config=smash_config)

To ensure fair usage within the free license limit, we’ve integrated product telemetry. This allows us to track the usage of the Pruna package and automatically send notifications when users are approaching their limit.

Currently, the limit is 100 free hours per month, and here’s the thing: it’s a soft limit. Why? Because we trust you (yes, you!) to play fair. We designed this system to avoid disrupting your production workflows. Only in cases of clear abuse would we reconsider this policy.

Finally, we’ve kept communication minimal and focused:

1. Community invite: A simple email encouraging you to join our Discord for assistance and collaboration.

2. Usage alert: A notification email when you’ve reached 80% of your monthly usage.

That’s it. No spam. No hard sells. Just tools and updates to help you get the most out of Pruna.

What’s next?

While the package is freely accessible, we’re constantly improving it. The next iteration will be on our documentation. We’re working to provide in-app assistance. This means you’ll be able to ask questions directly while reading the docs or even hop into Discord for real-time support.

Next up: AWS Marketplace integration! We know cross-platform compatibility is critical, and we’re working to make Pruna available wherever you need it. This is all part of our broader vision: making AI more accessible, one optimization at a time. So, go ahead, give Pruna a try, and let us know what you think!