JevMade hello@JevMade.com
← Back to guides

JevMade field notes / Hands-on tutorial

Compare local decision backends in the OpenJev workbench

A local workbench tutorial that starts the web application, optionally serves Mapika/decider on a CUDA GPU, and connects the UI to a decision backend for typed experiments.

Original by SiliconLabAIEvaluationGitHub README and appSource reviewed

Before you dive in

What you’ll find in the original

  1. Start the Node workbench first to explore its request and result interface with supported backends.
  2. For the Mapika path, install and serve `decider` separately; the documented 2B setup needs roughly 4 GB of CUDA memory.
  3. Point the workbench at the served endpoint and compare typed decisions rather than conflating the UI with the inference runtime.
Worth knowing

Requires Node.js; the documented Mapika backend additionally requires a CUDA GPU. This is an independent workbench, and its local backends are not represented as Jev-quality equivalents.