01 / DATASETS
A useful model starts with useful examples.
Plan JSON or JSONL collections around instructions, context, and expected outputs. Include the cases that reveal where your current model struggles.
instruction → context → expected outputDEEPDENSE / PRODUCT
A workspace for
your adaptation loop.
Keep the task, examples, and evaluation plan in one place.
Build a clearer path from a general model to your domain.
02 / EXPERIMENTS
Make every decision traceable.
Capture the base model, dataset, adaptation method, and settings. A reproducible configuration makes comparison easier when execution becomes available.
dataset + model + method + settings03 / EVALUATIONS
Look past a single score.
Compare expected and actual outputs using exact match and token overlap. Inspect mismatches and include baseline outputs to understand the difference.
baseline → comparison → failure review04 / DEVELOPER TOOLS
Explore before you integrate.
Browse the preview model catalogue, compose requests in the API Playground, and explore the API Keys screen. Execution and key creation are credits-gated.
models → playground → API keysCURRENT RELEASE
Explore the product.
Know the boundaries.
The workspace is in early access. Accounts and existing saved data remain accessible. Creating datasets, experiments, evaluations, API keys, and running model requests currently returns “Not enough credits.”
Model entries are previews. Provider training, inference, credit purchases, and billing are not available.
Explore developer tools