Explore the technical details
Ask your own question
This independent Jev Playground lets you test your own text and one question without supplying a TypeSafe API key. Choose Yes/No with your own rules, or Choice with 2–6 options that you name and describe.
This site’s free demo allows 10,000 characters in total across text, question, rules, option names and descriptions (Unicode code points). This is not Jev’s context limit.
These are model probabilities, not measured accuracy or a guarantee. Yes and No describe your rules; choices are relative to the options you supplied.
Change one detail in the text or one option, then run the question again to compare the responses. These small comparisons help you refine the task; they are not an accuracy benchmark.
Links from task guides load an editable template with one question. Feedback classification uses one Choice question; the writing template checks absolute promises; the reply template checks explicit requirements. These are smaller than the three-question illustrated scenarios described here. Loading a template does not run it or prefill a model result.
Live checks may require human verification and can pause when site quotas, the demo budget or service availability prevent a request. This independent demo does not grant access to your own TypeSafe account or API.
What are we asking?
These example outcomes were written to explain the workflow. They are not recorded Jev responses, and changing text does not run a model in illustration mode.
Sort customer feedback
Turn an incoming message into a suggested destination, without writing a reply for the customer.
- What kind of feedback is this?
- Does it describe an explicitly time-sensitive problem?
- Is essential information missing before someone can follow up?
The category suggests where to start. Urgency and missing information are separate questions.
Suggest a queue for a person to review. Keep account checks and any refund decision outside the model.
Check product writing
Ask a few clear questions about a draft before deciding what to edit.
- Does the text repeat a claim without adding information?
- Does it explain what happens after the reader’s action?
- Does it make an absolute promise without stated limits?
Each check answers its own question. A high score on one check is not an overall quality grade.
Return to the draft and decide which sentences to revise. Jev does not rewrite them or prove who wrote them.
Check whether a reply answers the question
Compare a request with a proposed reply before a person sends it.
- Does the reply address all explicit parts of the request?
- Does it state a concrete next step?
- Does it add a commitment that the request did not establish?
Covering the request, naming a next step and making a promise are different checks. A next step is not needed for every conversation.
Have the reply’s author review the missing points and verify any commitments. Nothing is sent automatically.
Your input stays out of our research
Running sends your text, question, rules and option names to TypeSafe. We do not store them on our server or publish them. Do not enter personal, confidential or sensitive information.
Saving is optional. One draft per language stays only in this browser for up to 7 days; expired drafts are removed when you return. Shared devices can expose saved drafts. Results are never saved to browser storage.
The counter uses Unicode code points; a combined emoji can count as several characters. Characters, UTF-8 bytes and tokens are different units. Jev’s official model limits are 64k tokens per request and 32k tokens for state plus the longest single question. The site’s 10,000-character allowance is not a 10k-token window or a guarantee that every input fits the model’s token limits.
A first-party cookie expires at midnight in Shanghai. Daily hashed security counters expire after two days, request status after seven days, and aggregate counts after 90 days. These records contain no input text or model results. After a successful human verification, a separate first-party cookie remembers that verification for up to 30 minutes. Changing networks or clearing cookies may require verification again.
We count page views, expanded cases, 30 seconds of visible reading, template loads, demo starts and completions, copied share links and continued reading. Events include the event type, page, language and task, plus only registered content, template and source-article IDs and a broad source category. They do not contain your text, questions, rules, results, raw referrer, full URLs or complete UTM parameters. A copied link does not prove a public share.
To connect a visit with a broad source category, this tab’s sessionStorage temporarily holds only the category and day. It resets on a new day and is cleared when the tab session ends. A source-article ID comes from a registered link and stays in page context rather than browser storage. We add no long-lived visitor identifier. This is separate from the optional local draft described above.
Keep exploring
When live checking is available, edit the input and choose “Check with Jev”. Otherwise, explore the illustrated workflow or use the official Playground.
Open the official Playground — TypeSafe account and product access required.
Sources & further reading
Official documentation and linked community reports support this guide. Community observations are attributed to their authors.
How we check our sources