Skip to main content
JEV is the evaluation workspace on HeroBoost Starter Agent chat. You create micro apps that score inventory (items you upload, load, or generate) against a rubric, review run history, and optionally run a rewrite loop to improve items that score below your threshold. Open chat → switch the side panel to JEV.
JEV is for the agent owner. Subscriber chat does not show the JEV panel. Optional API access uses keys with scopes such as jev_read / jev_run when you enable them for your workflow.

​
Words on the panel

​
Typical flow

1

Open JEV on your Starter Agent

From My Agents, find the agent → click the chat iconChat icon on My Agents — opens Custom Chat UI(to enter Custom Chat UI). In the header, choose JEV so the side panel shows the JEV workspace.
2

Create or select a micro app

Create a micro app, or select an existing one. You can rename or delete apps you no longer need. The selected app is remembered for this agent on this browser.
3

Build inventory

Add items by upload, load, or generate, depending on what your micro app supports. Confirm the inventory list before you run.
4

Set questions and rubric

Edit questions and rubric dimensions so a run knows how to score. Keep dimensions clear enough that results stay comparable across runs.
5

Run evaluation

Start a Run. When it finishes, open history / results. You can insert a summary into the chat composer when the panel offers that action.
6

Optional — rewrite loop

For items under your threshold, use the rewrite loop: revise those items into a new inventory, then run again. Auto mode can stop early when improvement stalls (product defaults exist for threshold and max rounds — adjust only if the UI exposes them).

​
Settings and access

  • Use JEV settings on the panel for micro-app options that belong with evaluation (not the global Configure tabs for model hosting).
  • If you issue API credentials for JEV, keep scopes limited to what you need (jev_read / jev_run).
  • Subscriber-facing JEV (if you turn on any “enabled for subscribers” option in agent config) is a separate product choice — default owner chat remains the primary path documented here.

​
Tips

  • Treat the first successful run as your baseline before you change the rubric mid-stream.
  • Prefer one micro app per evaluation purpose so history stays readable.
  • Pair JEV digests with Automation when you want scheduled follow-up — use Send to… only for destinations the UI lists.

​
Screenshots

UI walkthrough for the same JEV flow — switch tabs in order (create app → inventory → questions → run → results → rewrite).
Create micro app modal with Name and Source type label fields

Create micro app — name and optional source type label