Skip to main content
Jev is TypeSafe AI’s decision model. It writes no prose: you give it some content and a typed question, and it answers with a probability distribution, so the result is a value your workflow branches on rather than a sentence something has to parse. It connects with a TypeSafe API key.

​
What you can do

Actions - pick one option from a set, rate against a rubric, answer a yes/no question as a probability, ask many questions about the same content in one call, and list the models your account can use. Triggers - none. Use it through the assistant in chat or as an action node in an automation.

​
Get your credential

1

Open the TypeSafe console

Sign in at console.typesafe.ai and go to API keys.
2

Create a key

Create a key and copy it. Usage bills to that TypeSafe account, charged per input token - answers cost nothing.

​
Connect

1

Open Integrations

In arg.ai, go to Settings → Integrations and click Connect on Jev.
2

Paste the key

Enter your API key in the API key field.
3

Save

arg.ai checks the key against your TypeSafe model list and connects it.

​
Use it

  • In chat: “Score each of these bug reports for severity with Jev and list the blocking ones.”
  • In an automation: add a Jev action node to classify an incoming message, then branch on the option it picked.

​
The three question types

  • Pick one option takes a map of option name to what that option means, 2 to 255 of them. The answer names one of your options and carries a probability for every one of them.
  • Rate against a rubric takes 2 to 10 level descriptions ordered low to high, indexed from 0. The score is probability-weighted, so it can land between levels - 1.4 means the content sits between your second and third level.
  • Answer yes or no returns a single number from 0 to 1. Threshold it in your own code rather than treating 0.51 as a yes.

​
Confidence, and asking more than one question

A pick-one and a rate answer each carry a confidence between 0 and 1, derived from how concentrated the probability is. Read it as a second axis: the answer says what, and the confidence says whether to act on it or hand the item to a person. Jev reads the content once and answers every question against it in parallel, so ask many questions at once is markedly cheaper and faster than one call per question - which makes speculative questions you may not end up reading nearly free. Ask everything you might want in one call and decide in your own code which answers matter.

​
Limits

One request covers 64k tokens of content plus all of its questions, and 32k for the content plus the single longest question. Input is text: a string, or a JSON object or array of text values. Pre-process images, audio, and binaries into text before sending them.