@jevkit/all

Tooling for building on TypeSafe's Jev, the first System One model. Jev returns typed answers and calibrated probabilities instead of generated text; jevkit is the ring of tools around that.

This package installs nothing of its own. It is a convenience umbrella that pulls in the whole set:

npm install @jevkit/all

| Package | What it does | | --- | --- | | jevkit-core | the .jevl record format, canonical request digests, question normalization, token budgets | | jevkit-lint | static linting of your questions before you pay for them | | jevkit-drift | detect when answers move under you | | jevkit-bench | score a labeled suite for accuracy and cost, and compare two runs | | jevkit-calibrate | reliability diagrams, ECE, Brier, and confidence thresholds from labeled outcomes | | jevkit-vitest | vitest integration |

Prefer installing only what you use. The umbrella exists so one install gets you the whole ring.

The unscoped name jevkit is unavailable: npm blocks it as too similar to an unrelated package, jev-kit, published two hours after these packages.

The same tools are on PyPI as jevkit-core, jevkit-lint, jevkit-pytest, jevkit-drift, jevkit-bench and jevkit-calibrate. See jevkit-py.

Unofficial and unaffiliated with TypeSafe. Not an SDK: TypeSafe ships its own.

MIT licensed.