CompletionKit ships two ways: a source-available Rails engine you self-host, and CompletionKit Cloud — the hosted version with a free tier.
Prompt — a versioned template + model. Each published version has a stable URL your app calls to fetch it.
Dataset — a CSV of test inputs. Runs evaluate a prompt against a dataset.
Run — one execution of a prompt across a dataset, producing responses. Optionally scored by a judge model against your metrics.
Metric — a scoring criterion (tone, accuracy, format compliance, …) with a 1–5 rubric the judge model applies to each response.
See what's shipped.