Jev for trust and safety teams: what it can do and what it costs

Jev, TypeSafe AI's decision model, answers a question about text with a value your software can use: one option from a list, a score, or a yes/no probability. That suits the small, repeated decisions in the middle of this work.

These are the jobs on this site that fit trust and safety teams best. Each has its own page with the questions to ask, the pitfalls and a worked example.

• Flag spam, abuse and prompt injection before content goes live. Jev works well as a fast first-pass filter: one request returns a probability for each policy you define (spam, abuse, prompt injection).
• Check every message and reply for jailbreaks and hazards before the LLM acts on it. One Jev request per message can check it against several hazards at once, such as jailbreaks, requests for harm, medical advice and self-harm, and rate how much harm answering would do.
• Tag ads by angle and hook strength, and pre-check them against policy. Jev can tag each ad by angle, rate hook strength and check it against your policy list in a single request, which makes it practical to label thousands of competitor or historical ads.

• Jev for content moderation
• Jev for LLM guardrails
• Jev for ad and creative classification

Jev lists $0.042 per million input tokens with output free, so the cost depends on how much text you send and how many questions you ask about it. At the typical volumes the use-case pages assume, the calculator estimates:

• Posts (content moderation): about 1.1¢ per 1,000, at 3 questions and about 150 tokens each.
• Messages (LLM guardrails): about 2.3¢ per 1,000, at 5 questions and about 300 tokens each.
• Ads (ad and creative classification): about 1.3¢ per 1,000, at 3 questions and about 120 tokens each.

• Cost calculator
• Jev pricing and real runs

A plain way to begin, in order:

• List your policies as separate yes/no questions, one per rule, in the words your policy uses.
• Run a labelled sample of past decisions through them and note where Jev and your moderators disagree.
• Set a threshold per policy, and send anything near the threshold to a person.
• Keep humans on final removals and appeals, as the use-case page advises.

• How to use Jev

Jev reads text only by default, so images, audio and video need another route, and it is a first-pass filter, not a final judge. TypeSafe's own guardrails example recommends checking both the user's message and the model's reply.

From the use-case pages, Jev is not the right choice when:

• You need legally defensible moderation decisions with written reasoning.
• You need the reason a message was blocked written out for the user. Jev returns probabilities, not explanations.
• The creative is image- or video-only with little text.

• What Jev can't do

This is a good fit for the first pass and a poor fit for the last word. The value is that each policy gets its own probability, so your team can choose different thresholds for different harms instead of one blunt score.

The figures above are list-price estimates, not measured bills, and the accuracy claims on the linked pages are the published runs' own. Test Jev on a labelled sample of your own data before you rely on it.

• What is a decision model?