Privacy preferences

Strictly necessary

Always on

Remembers your choice on this page. Stored in your browser as gb_consent for up to 6 months. If you switch the site to dark mode, that choice is kept as gb_theme. Nothing is sent to a server.

Anonymous analytics

Helps us count visits. At most once every 24 hours we send one ping to our own telemetry server, using the same anonymous approach as the WeightsLab package. No advertising, no third-party trackers, no cross-site tracking.

Exactly what is sent
uuid
A random ID created in your browser (gb_tid). It is not linked to your name or email.
event, version
website_visit and website, so website visits can be told apart from WeightsLab package pings.
os
Operating-system family only, for example Windows, Darwin, Linux or Android.
tz
Your time zone, for example Europe/Paris.
On our server
Your IP address is used to estimate country and city. Only a salted hash of it is kept, never the raw address.
GitHub star count
With analytics on, your browser also asks api.github.com for the public star count of our repository, so GitHub sees your IP address. With analytics off, the page shows a fixed number and makes no request to GitHub.

Turning analytics off deletes gb_tid and stops all pings.

Cookie policy · Privacy policy

About Graybx

Why we started WeightsLab.

Graybx builds WeightsLab, an on-prem training workspace for observing, diagnosing and intervening on model training data.

We have worked together for over a year. One of us came from frontier model training, the other from medical AI, where data and explainability matter, and we hit the same problem in both: when a model fails, nobody can say which data caused it, and every fix is another full run. WeightsLab is what we built to deal with it.

Alexandru-Andrei Rotaru

Alexandru-Andrei Rotaru

Co-founder & CEO

Former Google AI engineer on Assistant and Gemini. Started WeightsLab after hitting these training problems first-hand inside a large research organisation.

Guillaume Pelluet

Guillaume Pelluet

Co-founder & CTO

PhD in explainable multimodal AI. A decade moving AI from research into CE and FDA deployments, and growing a medical-imaging scale-up team approximately tenfold.

Supported by

  • NVIDIA Inception Program
  • TUM Venture Labs
  • France 2030
  • Scaleway for Startups
  • National Edge AI Hub
  • OVHcloud

FAQ

Questions we hear a lot.

Something missing? Write to hello@graybx.com.

What is WeightsLab?

An on-prem training workspace that hooks into your PyTorch loop. While the model trains it shows per-sample signals, helps you work out what is going wrong, and lets you act on it without restarting: drop, retag or re-annotate samples, steer the run.

How is it different from experiment trackers and data-curation tools?

Trackers report on runs, and curation tools score datasets after the fact. WeightsLab sits inside the run: it connects model behaviour to individual samples and lets you change them while training is still going.

What do I need to get started?

A supervised-learning dataset, a base model and a PyTorch training script. Run pip install weightslab, wrap your data, model and loss, then launch the interface locally.

Where does it run, and who owns my data?

Always on-prem, on GPUs you manage, for every customer. You keep ownership of your data and your models, and training stays on your infrastructure.

Which LLM does the built-in agent use?

It uses the LLM you pick. The agent runs on OpenCode, so it works with any LLM provider OpenCode supports, such as OpenRouter, Anthropic, OpenAI or Google, and with models you host yourself through Ollama, LM Studio, llama.cpp or any OpenAI-compatible endpoint. Your provider credentials stay in OpenCode, never in WeightsLab, and with an on-prem model your sample-level data stays on your machine.

Which models and data types are supported?

PyTorch training loops, with image, video, text, tabular and LiDAR data. Classification, detection and segmentation are covered by our examples, and computer vision, Physical AI perception and medical imaging are where we are deepest. Training or fine-tuning a language model? Tell us about your setup and we'll walk through it with you.

How much does it cost?

WeightsLab is free for students and for individuals working on personal projects. Companies of any size license it per seat. A seat covers one user and includes support, updates and the custom features your team needs. WeightsLab runs on your own infrastructure either way. Talk to us for the terms that apply to your team.

See it on one of your runs.

Try the sandbox in your browser, or book 30 minutes with us and bring a real run.

Book a demo Try the sandbox

Join the early-access programme

Tell us what you train and we'll get back to you with a demo and licence terms.