Discern

Checked, not fully reviewed. We built it and ran its tests — 64 pass. No Jev calls.

Effect v4 ships Decision and DecisionModel, a provider-neutral vocabulary for semantic judgments. Discern sits on top as the semantic counterpart to Match: you write patterns over the evidence a DecisionModel returns, attach policies, and register procedures that can be routed to by meaning. The point it makes well is that an uncertain answer is a case to handle — “maybe” is a branch in the type, not a probability you forget to check.

Discern makes no network call of its own. The provider is a Layer; the README’s example wires TypeSafe’s Jev through @effect/ai-typesafe, and anything else that implements DecisionModel drops in the same way, which is what makes it testable without a key: the suite runs against a scripted model, so a policy’s branches can be exercised without spending a request.

MIT, 15 stars. Narrow by design — useful exactly to people already writing Effect.

See how it compares with other tools in Best Jev tools, tested hands-on.

Review updated Oct 4, 2026. Numbers quoted from the project are its author's own; we don't publish our own measurements of Jev.