Question 1

Do three questions in one call cost the same as three separate requests?

Accepted Answer

Barely more than one, per the README: ask() sends every question about one state in a single call, the questions are answered in parallel and each costs only its own tokens, so three questions about one email come to roughly the price of one call instead of three round trips. The trade-off is on the other axis — askEach() across many rows is sequential by design, so each result stays attached to its row and each failure stays attributable, and parallelism belongs in your queue or worker pool.

Question 2

How do I split work between automatic routing and human review?

Accepted Answer

On confidence. uncertain(threshold: 0.8) returns the names of the fields that fall below the bar, which you flag for review, while a choice at or above 0.8 can be routed straight away. Note the asymmetry the README calls out: a Noul has no separate confidence — its probability already carries it, 0.5 meaning no information and certainty growing towards either end — and confidence() returns the distance from that midpoint so one interface covers all three question types.

Question 3

What happens if the API is rate limiting or overloaded while a job runs?

Accepted Answer

RateLimitException on 429, OverloadedException on 529 and ConnectionException are retried automatically with exponential backoff and jitter, honouring Retry-After, and you can tune it with RetryPolicy(maxAttempts: 5, baseDelay: 0.5, maxDelay: 8.0). A 401 AuthenticationException or a 422 ValidationException is never retried, because retrying cannot help. In a queued job pass RetryPolicy::none() and let the queue's own tries and backoff own the retry.

Question 4

Can I test the Laravel integration without a key or network calls?

Accepted Answer

Yes. Bind a Jev built with your own PSR-18 client plus PSR-17 request and stream factories and nothing leaves the process: $this->app->instance(Jev::class, new Jev('test-key', http: $fakeClient, requests: $psr17, streams: $psr17)). For config caching, put JEV_API_KEY in .env but read it through config/services.php, bind the client once as a singleton in AppServiceProvider::register(), then type-hint Jev in controllers, jobs, commands and listeners.

Question 5

What PHP and framework requirements does the SDK have?

Accepted Answer

PHP 8.1 or newer and any PSR-18 HTTP client. It is framework-agnostic and works in plain PHP, Laravel 8.x on PHP 8.1+, Symfony, or anything with Composer.

Question 6

How do I keep answers stable across deploys?

Accepted Answer

Pin a model version instead of using jev-latest, for example withModel('jev-1.13.0') or the model argument to the constructor. The README recommends pinning so answers stay stable.

Question 7

Why does askEach run its requests sequentially?

Accepted Answer

So each result stays attached to its row and failures are attributable. The README says to put parallelism in your queue or worker pool, where you also control how much of the API you use at once.

Jev Php SDK

About this project

Highlights

Questions about this entry

More Jev resources like this

From the Jev guides

From the Jev community