Deployment

Where it runs.

The demo's hosted worker, and a lender's own environment.

In the demo

The demo is hosted. Labelled runs are scored by a hosted GPU worker. A Day-Zero run on your file is computed by a hosted CPU worker, which sends the model provider the column names and descriptions, the lines about the lender, and statistics and eight rows of the file, with the consent you give on the form. The seven screens on a public book compute in your browser from the card and the book.

In production

zypl.Foundation runs inside the lender's environment, on server CPUs, and no borrower record enters its pretraining:

C1zypl labs research paper (under review), §3 and appendix A

Pretrained on synthetic credit economies; no borrower record enters pretraining.

Own-brand instances add continued pretraining on the bank's data, inside the bank's environment. Synthetic-only pretraining and bank-local serving are separate properties.

C5zypl labs serving report v3.1, tables 5 and 17

~50 ms

Scores an application in about 50 ms (54 ms median, 57 ms p95) on a 16-thread server CPU and about 20 ms on a GPU, matching the research model's predictions.

After a one-time portfolio encoding (about 90 s on CPU, 5 s on GPU, can run offline); 12,000-row context; one server.

C7zypl labs serving report v3.1, table 24

4–8

Runs on commodity CPU servers: 4–8 cores and about 7 GB RAM per worker.

Memory scales with portfolio size and member count. This is production sizing, not the 16-thread host the latency figure was measured on.

The demo is not a production instance

The demo exists to show the product on public data and on files you bring for evaluation. Its storage is not meant for borrower files, and its accounts are for evaluation. A production instance is set up inside the lender's own environment.