GPU
Auspex RT
Bounding-box detection, trained from scratch on your labels.
COCO JSONYOLO
pip install auspex-rt==0.2.0
For Kaggle notebooks and scripts
Write it here.
Train it on Kaggle’s GPUs.
Caudex Studio is a browser workspace for notebooks and scripts. Press Run and it pushes your code to Kaggle, streams the log while the model trains, and hands you the checkpoints when it’s done.
Nothing to install. It runs on your own Kaggle account and its weekly GPU hours.
Works withPython scriptsJupyter notebooksCOCO JSONYOLO data.yamlCVAT 1.1 XMLKaggle datasets
Features
Everything between an idea and a trained checkpoint
The whole Kaggle loop, from edit to push to watch to download, in one tab.
Scripts and notebooks, edited in place
Write a .py file or work cell by cell in an .ipynb, with Python highlighting, autosave and Ctrl+S. A template’s settings cell shows up as a simple form.
Run on Kaggle in one click
Choose CPU, a T4 or P100 GPU, or a TPU, then press Run. Your code goes to Kaggle as a new notebook version; nothing trains on your laptop.
The log, while it trains
Kaggle’s log streams into the page as it is written. Lose the connection and it resumes where it stopped, without repeating a line.
Outputs, straight from Kaggle
Checkpoints, metrics and plots are listed by folder with their real sizes, and download directly from Kaggle, even a 191 MB best.pt.
Your datasets, attached
Search Kaggle for a dataset and attach it to a project. It is mounted under /kaggle/input when the run starts.
Quota and history at a glance
See how many GPU and TPU hours are left this week, and every past run with its status, log and outputs.
How it works
From sign-in to best.pt in three steps
- Step 1
Sign in and connect Kaggle
Sign in with Google or GitHub. Then, on kaggle.com, open Settings → API, create a legacy API key and paste the kaggle.json it downloads.
{ "username": "you", "key": "••••••••" } - Step 2
Pick a template, attach your data
Start blank or from an Auspex model, attach a Kaggle dataset, and set the epochs and image size in the settings form.
auspex-rt · GPU (T4) · /kaggle/input/street-signs - Step 3
Run, watch, download
Press Run on Kaggle, follow the log as the model trains, then download best.pt from Outputs.
runs/rt-run1/best.pt · 191 MB ↓
Templates
Start from a model that already trains
Each Auspex template installs its pinned package on Kaggle, finds your dataset under /kaggle/input, picks mixed precision for the GPU it lands on, and copies what matters into artefacts/.
GPU
Auspex Seg
Instance segmentation: a polygon, class and box for every object.
COCO polygons
pip install auspex-seg==0.2.1
GPU
Auspex Vision
Boxes, polygons, keypoints, polylines and tags, all in one model.
COCO JSON
pip install auspex-vision==0.7.0
GPU
Auspex Cursus
Polyline detection: an ordered sequence of points per instance.
CursusJSONCVAT XML
pip install auspex-cursus==0.8.2
GPU
Auspex Engine
An auto-annotator: one specialist per annotation type, bundled into one model.
COCO JSONYOLOCVAT XML
pip install auspex-engine==0.3.1
CPU
Blank script or notebook
Trains nothing: prints the environment and writes proof.txt, to check the whole loop at zero GPU cost.
No dataset needed
Start here on your first run
Security
Your Kaggle key stays yours
Your key is checked with Kaggle, then encrypted before it is stored, with a secret kept apart from the database. It is used only to act for you on Kaggle.
- Checked with Kaggle once, then encrypted with AES-256-GCM before it is stored. A copy of the database alone holds no usable key.
- Each account sees only its own projects, files and runs.
- Revoke the key on Kaggle and it stops working here straight away.
- Your code runs on Kaggle, in your account, on your own quota.
You sign in with Google or GitHub
Firebase confirms who you are. We never see that password.
You paste your kaggle.json
Your Kaggle username and API key, typed or pasted.
Kaggle confirms it, once
A wrong or revoked key is refused before anything is saved.
Encrypted before it is stored
AES-256-GCM, with a secret kept apart from the database.
Pricing
Free for 7 days, then one plan
No card to start. After the trial, pay monthly by card, or 30 days at a time in taka. Kaggle’s GPU hours stay free: the plan pays for Caudex Studio, not for Kaggle.
Free trial
7 daysfree
- Starts when you sign up
- Everything the paid plan includes
- No card, so nothing is charged when it ends
Pay by card
$5a month, in USD
- Renews every month until you cancel
- Cancel any time and keep access to the end of the paid month
- Paddle, our reseller, handles the payment, sales tax and VAT
Pay in taka
৳650for 30 days, in BDT
- bKash, Nagad, Rocket or card
- Paid through SSLCommerz
- One payment, no auto-renewal
When the trial or paid time runs out, nothing is deleted: you can still sign in and open your projects, code, runs and outputs. Starting runs, creating projects and uploading datasets wait until you pay.
The terms and refund policy have the details.
FAQ
Questions, answered
Do I need to install anything?
No. You need a Kaggle account and an API key. Editing happens in the browser and training happens on Kaggle.
What does a run cost?
Runs use your own Kaggle quota. Kaggle includes free GPU and TPU hours every week, CPU runs don’t touch the GPU hours at all, and the dashboard shows what is left.
What am I paying for, if Kaggle is free?
Caudex Studio itself: the editor, one-click runs, the live log and the outputs. Kaggle’s GPU and TPU hours stay free and come from your own Kaggle account; the plan doesn’t add to them.
What happens when my free trial ends?
Nothing is charged, because the trial never asks for a card, and nothing is deleted. You can still sign in and open your projects, code, runs and outputs. Starting runs, creating projects and uploading datasets wait until you pay, and the same goes when paid time runs out.
Which hardware does my code run on?
The runtime you pick: CPU, an NVIDIA T4 or P100 GPU, or a TPU VM v3-8, as Kaggle provides them.
Where do my projects live?
In Caudex Studio, under your Kaggle account. Each run also pushes the code to Kaggle as a new version of a notebook in your account, so it is there too.
Can I stop a run from here?
Not yet. “Stop watching” stops following a run here, and every run links to Kaggle, where you can cancel it.
Is Caudex Studio made by Kaggle?
No. It is an independent tool that talks to Kaggle’s public API. It is not affiliated with or endorsed by Kaggle.
Your next training run is one click away.
Sign in with Google or GitHub, connect your Kaggle account, pick a template and press Run.