Build with Jev.
Start with one task.
Jev classifies text, scores candidates and answers yes/no questions. Learn the API, run a Python example and check the results.
{
"model": "jev-latest",
"state": "Where is my order?",
"questions": {
"team": {
"type": "choice",
"instructions": "Choose a support team",
"criteria": {
"shipping": "Delivery and tracking",
"billing": "Payments and invoices"
}
}
}
}Watch Jev on Android.
Play the Android-agent demo here. For official comparisons, Doom and Wikiracing, open the video collection.
Explore four video demos →By droidrun / Mobile Jev contributors. The recording stops at payment selection.
Use the play control to watch here, or open the original video above.
Browser actions, email routing, coding-agent checks and more, with original projects and evidence notes.
11 official articles, patterns and evaluations, with summaries and related guides.
What are you here to do?
Choose a task to try
Route a support request
Suggest a support team, or ask for review when the request is unclear.
Classify support tickets
Compare expected labels with predictions and inspect errors.
Rerank search results
Score candidate passages and check the top two.
Flag possible threats
Try a review policy with two example cutoffs.
The Python demos use supplied answers. Read what was tested before interpreting their results.