> ## Documentation Index
> Fetch the complete documentation index at: https://docs.connie.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Jev by Typesafe

> Qualify information against your own options, in a Workflow step or as an action column.

**Jev by Typesafe** qualifies information: you give it some text and a list of options, and it answers with the option that fits, or with every option that applies. Use it whenever you want to sort, score or check something into categories you define: lead fit, intent, segment, priority, sentiment, or a yes/no check. Jev is a decision model by TypeSafe. It does not write text; for that, use an AI Agent step or column.

You can add it as a **Workflow** step (**Add an action**, then **AI**) or as an **action column** on a List (**Add a column**, then **AI**). Both open the same settings.

## Settings

* **Information to qualify** — the text Jev judges. Insert a Field or an earlier step's output with the variable picker, for example a company's description.
* **What to decide (Optional)** — the question Jev answers, such as "Is this company a fit for us?" or "Which buying signals does this show?".
* **Output** — **Single select** picks the one best option. **Multi select** returns every option that applies.
* **Options** — the answers Jev chooses from. Each has an **Option** (the answer as it should appear in the result, for example **Qualified**) and a **Criterion**: what makes that option the right one, for example "B2B company with 50+ employees and its own sales team". Jev judges the information against the criterion, so describe each option clearly. An option without a criterion is judged by its name. Click **Add** for another option; empty rows are ignored.

Single select needs at least two options, multi select at least one, and you can use up to 50.

## What you get back

* **value** — the chosen option (single select), or the first option that applies (multi select; empty when none applies).
* **values** — every option that applies.
* **confidence** — how sure Jev is, from 0 to 1. For multi select, the lowest probability among the options that apply.
* **probabilities** — Jev's probability for each option.

In multi select, an option applies when its probability is 0.5 or more. Use **probabilities** in a later step if you want a stricter or looser threshold.

## Pricing

Each run is charged for the size of the information and options Jev reads (its input); the answer it returns is free. The charge is TypeSafe's price for that input plus 5%, converted to credits and rounded up to the nearest 0.01 credit. A typical run, a paragraph and a few options, costs **0.01 credits**; 100,000 tokens of input (a long document) costs about 0.27 credits.

The step's settings show this as an estimate beside the name, and on the **Run Test** button, worked out from what you have typed and updated as you type:

* **\~0.01 credits** — the estimate for one run (or one Record, in an action column). Hover it for the estimated input size.
* **from \~0.01 credits** — the information or options include an inserted Field or step output. Its length isn't known until the run, so the estimate counts it as empty and a run costs at least this. In the **Test** panel, the values you enter for the test are counted instead.
* **from 0.01 credits** — in **Add an action** and **Add a column**, before the step has any settings.

Multi select asks Jev one question per option, so the same options cost a little more than in single select. The estimate is within a few percent of what Jev reads; you are charged for what it actually reads. After a test run, the **Test** panel shows what that test was charged.

A run that fails before Jev answers is not charged. What each run was charged appears in **Logs** and, for a column, in the record's side panel.

## When something goes wrong

* **"Add at least two options to choose from."** — single select needs two options with a name.
* **"Jev by Typesafe could not read this request."** — check the information and options; the request is not retried.
* **"Jev by Typesafe is receiving too many requests"** or **"temporarily unavailable"** — the run tries again by itself; nothing is charged for the failed attempt.
* **"The information is too long for Jev by Typesafe."** — shorten it or map a smaller Field.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.