Playground/Conversation lab
API ACTIVITY

Usage & budget

Your TypeSafe API key

Import your key to override the shared server default.

Saved on this browser until removed or site data is cleared. Masked on screen; sent only with Jev requests through this app. The saved value is never displayed.

Connecting
Conversation lab

Give the right message a reply.

Paste a conversation and see where a response would matter most.

Preview only · no messages sent
WORKSPACE GUIDE

Conversation lab

Rank messages and find the person your bot should answer next.

MessagesReply policyRecipient

What you provide

A conversation transcript and experiment choice: select a reply recipient or compare how a changed detail affects the final message.

What happens

Reply selection evaluates each speaker’s latest message; comparison modes evaluate the final message. Jev returns reply probabilities, and the configured probability threshold controls selection.

What you get

A proposed recipient, the other candidates, and supporting scores. Changing the threshold recalculates the displayed decision locally.

Walk through a run

  1. 01

    Set the conversation

    Load a sample or paste messages, then choose the experiment.

  2. 02

    Choose a recipient

    Run Jev to rank messages against the reply policy.

  3. 03

    Inspect the ranking

    Review the selected message, other candidates, and the reply-probability threshold.

Try this

Load the sample, pick a recipient, then raise the threshold and watch when the selection becomes uncertain.

Execution limits

Live classification only. The lab does not write or send replies.

Explore the example. Check the evidence.

Your conversation

Paste from Discord, or use Name: message.

Parsed messages 3
Val

Do they have a desktop app?

Tyler

Can you prototype the codebase search flow? I need exact context, line-by-line search, and AST support.

Morgan

That sounds great 👀

Policy & evaluation

Labels are scored in A/B and final-message modes.

Up to 3 requests in parallel

The reply decision

MessagesReply policyRecipient

Who deserves the next reply?

Run the conversation to compare reply probabilities and identify a recipient.

More repliesMore selective
Frame evaluation

Valid labeled runs in this tab. Repeated runs count again; this is not a benchmark.

With context: 0/0

Latest message only: 0/0