Crasis Studio · Hosted pipeline

Describe a task.
Receive a specialist.

No GPU. No API keys. No code.

The full Crasis pipeline, hosted. Tell us what you need classified. We handle the training data, the GPU, and the ONNX export. You get a model that runs locally forever.

Join the waitlist

Studio is in development. Be first to know when it launches.

No spam. One email when Studio launches.

how studio works

The pipeline,
without the pipeline.

Everything the CLI does — data generation, training, eval, export — handled for you in a browser.

01 — describe

Write a task description

Tell Studio what you want classified in plain English. What does a positive example look like? What should it ignore? No code, no YAML — just a description.

02 — build

We run the pipeline

Studio generates synthetic training data using distillation-licensed models (enforce_distillable_text: true), trains on cloud GPU, and validates against your quality bar. Takes 4–30 minutes.

03 — deploy

Download and run forever

You receive a single ONNX file. 4–11MB. Runs anywhere — laptop, Raspberry Pi, edge device. No API key at inference time. No ongoing cost. Ever.

Pay for the build.
Not the queries.

Pricing is based on training complexity — task type, sample count, and compute time. Quoted upfront before you commit. No subscription for the build itself.

~$19 binary classifier
~$49 multiclass / extraction
custom enterprise / private registry

Exact quote generated from your spec before payment.

Want to run the pipeline yourself?

Crasis CLI is open source and free →