TOOLS / JEV WALKTHROUGH

Jev AI for B2B marketing

Jev is TypeSafe's model for structured judgments about text. This demo shows how an application can use those judgments to select context before an LLM researches an account or develops content.

Illustrative demo

Explore a fictional ERP consultancy selling to manufacturers and distributors. Select an example, then press Play or Next step. Click a card to inspect its evidence and authored judgment.

Business context

CLARIFICATION LAYER

Jev

Ready to evaluate

Selected context

0

LLM workspace

DEEPER ANALYSIS + WRITING

Needs a closer look

Uncertain inputs stay available for human review.

SPACE TO PLAY · → TO STEP
Fictional inputs and prewritten results. No live model calls.Jev: $0.042 / million input tokensPublished rate · checked September 29, 2026

BEHIND THE VISUAL

How Jev selects context

Ordinary code handles known rules, exclusions, and suppression before semantic checks. Software supplies the business context and specific questions. Jev returns structured judgments. Software uses those judgments to select context or flag items for review; an LLM can then analyze or write from that context.

Choice

Select a defined category, such as fit, no fit, or insufficient evidence.

Score

Rate a specific dimension, such as relevance, against a defined rubric.

Noul

Estimate the probability that a condition holds, such as a relevant business change.

Jev can also check LLM outputs against source text. A source-support check assesses what the source supports; it does not establish whether the source itself is true.

Message relevance is an assessment against buyer evidence. Actual campaign results establish response and conversion performance. Confidence is not a guarantee of correctness. Jev judgments never authorize outreach or publication.

Filtering can reduce downstream work. Net cost savings depend on the workload, review needs, and the models used. This demo has no measured savings or live Jev scores.

Use cases ↗Routing ↗Source checks ↗

How to add Jev to your stack

Start with one decision your application needs to make. The demo above uses fictional, prewritten examples. These steps describe how to try the real hosted service.

  1. Try Jev first. Open the TypeSafe Playground. You need a TypeSafe account and appropriate access to use it. Try a few examples before connecting an application.
  2. Choose one task. For ICP fit, supply company evidence, define buyer criteria, and ask whether the company matches those criteria. Define outcomes for fit, no fit, and insufficient evidence. Keep missing evidence separate from a negative answer.
  3. Connect it to your application. Use an SDK or the official API. The SDK connects to TypeSafe's hosted API; it does not install Jev's model locally. Calls require account access and may incur usage charges.
Optional SDK installation commands

Use Python 3.10 or newer.

pip install typesafe-sdk

Read the Python SDK instructions

  1. Keep your application in control. Configure TYPESAFE_API_KEY in your server-side environment. Keep it out of browser code. Your application applies known rules, handles uncertainty, and routes selected context to an LLM. Test against representative examples and review errors before production use.

This page does not collect credentials or execute these commands.

Open the official Jev quickstart →

Seven applications to explore

These are proposed B2B marketing workflows based on TypeSafe's documented classification, ranking, and verification capabilities. The examples show where a narrow judgment could help, rather than measured client results.

ICP identification

Select companies whose documented industry and operating problems match your buyer criteria. An LLM can research the selected accounts. Fit does not establish budget or purchase intent.

Message relevance

Check whether a message addresses a documented buyer problem. An LLM can develop revisions from that evidence. Relevance scores do not predict response or conversion.

Content checks

Flag passages that are vague, irrelevant, or unsupported by the supplied source. An LLM can revise them. Source support does not establish that a source is true.

Buying signals

Assess whether an announcement describes a business change relevant to your offer. An LLM can investigate the account. A relevant change does not prove purchase intent.

Customer research

Identify objections and pain points in supplied interview excerpts. An LLM can synthesize the selected evidence. The excerpts must be representative enough for the question.

Context selection

Rank source passages for a specific assignment so an LLM can work from focused context. Keep the original sources available and check what filtering leaves out.

Reply triage

Separate questions, interest, and opt-outs for review. Known opt-outs remain blocked in both routes. Application rules and human approval control suppression and any outreach.

What this demo can tell you

Compare the same inputs with a Jev clarification step or a direct route to an LLM. Click a card to see the evidence, question, and illustrative judgment. Neither route is a benchmark: every company, input, judgment, and output is fictional and authored in advance.

Jev does not replace source verification, suppression rules, permissions, or human review. Confidence is not a guarantee of correctness. A live workflow needs an evaluation set, uncertainty handling, and monitoring appropriate to its task.

Context selection can reduce downstream work. Net savings depend on input volume, review effort, and model costs. This demo measures neither savings nor model quality.

TypeSafe's model pricing currently lists Jev at $0.042 per million input tokens. Reviewed September 29, 2026. Check the official page before budgeting.

Official documentation and related reading