seabiscuit

Seabiscuit Cloud

Everything around the language.

Hosted runtime and API, managed training, a data store, a live-graph code editor and drag and drop graph editor, git-backed deployments to a mobile app, the template for said mobile app. Available through early access.

REQUEST EARLY ACCESS →TALK TO US →
  1. 01 · WRITEWrite with a live graph
  2. 02 · TRAINTrain on your data
  3. 03 · DEPLOYDeploy and roll back
  4. 04 · CALLCall it from anywhere
EDITOR

An editor that draws as you type

Write policies in the browser with the graph alongside. Every keystroke recompiles, so the graph and the diagnostics are never out of date.

  • A live graph that updates as you type, with fn calls folded into groups
  • An Inference view that hides the training-only parts of the graph
  • A resizable split between code and graph
  • rustc-style diagnostics inline, with the fix in the help
TRAINING

Managed training

Declare the optimisers in the policy and start a run against a dataset. We run it, keep every artifact, and show you how it went.

  • Built-in optimisers such as Adam and CMA-ES, plus closed-form delegate fits for nodes like linear
  • Holdout, k-fold and leave-n-out splits, grouped so repeat measurements never straddle train and validation
  • Out-of-fold predictions, metrics and optimiser panels for every run
  • Evaluate, sweep and compare runs side by side
DATA

A data store for entries

Inputs and labels live together as entries, organised in folders, with point clouds and other files stored alongside.

  • Folders with schemas that check the inputs and labels you upload
  • CSV import and export for tabular data
  • Tags for slicing datasets across folders
GIT

Git-backed policies and deployments

Every policy lives in a real git repository. A deployment pins a commit and the exact trained values that go with it.

  • History, diffs and commits for every policy
  • “Move deployment marker to here.” on any commit
  • One-click rollback: code and trained tensors restored together
  • Locks pin every import, so production runs are reproducible
RUNTIME

Hosted runtime and API

Deployed policies run on our infrastructure. Call them from your own apps and get typed outputs back, distributions included.

  • A REST API for runs, data, tensors and training
  • Per-node results and failures, with healthy outputs still returned
  • One runtime for running and training, so what you validate is what you ship
CAPTURE

A mobile capture app template

A ready-made phone app for collecting scans in the field and running your policies on them, which you can brand as your own.

  • LiDAR and camera capture into the data store
  • Runs deployed policies on new scans
  • Your name and colours on it

In the field

From a phone scan to a policy run.

The capture app template is how Equimorphic collects horse scans today: capture with the phone's LiDAR or camera, upload into the data store, and run the deployed policies on the result. Label those entries and they become your next training set.

Your users get your name on the app. You get the scans in the same place you train.

The capture app showing a LiDAR scan of Gerald, Equimorphic's test horse, with the policies that can run on it.

Access and pricing

Early access now. Prices with the beta.

We're onboarding a small number of teams by hand while Seabiscuit Cloud gets ready for everyone. Paid plans and their prices will be published with the public beta. Until then, tell us what you're building.

Early access

For teams building on Seabiscuit now. The full cloud, set up with us, with a direct line to the people who wrote it.

Enterprise

Custom nodes for your data, white-labelled apps, private deployment on your own infrastructure, and terms set for your scale.