jev-feels

Checked, not fully reviewed. We ran its own suite and the documented quickstart, and measured what a call sends. We made no Jev calls and used no API key.

feels :body, :urgent turns a named question into email.feels?(:urgent); decide and score cover the other two answer types, and validates_feeling puts a threshold into Active Model validation. The declaration binds a question to a field, which is the part that reads well: the question lives next to the data it asks about rather than in a service object three files away.

179 examples pass on Ruby 3.4, with the one live-API example correctly pending without a key. The state it sends is the attribute’s text and nothing else.

The cost shape is worth knowing. We measured three predicates on one object called one at a time: three requests, one question each. Asking the same predicate twice on the same object asks twice — there is no per-instance memo, so a view that calls feels? in a loop pays each time.

Jev.measure exists for that, and batches several questions into one request — but the README’s batch example passes a record (Jev.measure(ticket)) where the method wants text, and raises ArgumentError: text must be a String. Passing the text works and gives one request for all the questions. We reported it; until it is fixed, reach for Jev.measure(record.body) { … } rather than several predicates in a row.

MIT, 19 stars, a week old. Pleasant to read, and the smaller cousin of Truffler, which stores answers and ranks in SQL instead.

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

Review updated Oct 1, 2026. Numbers quoted from the project are its author's own; we don't publish our own measurements of Jev.