Skip to main content

Introduction

A guide to the Multiverse Foundry Beta: importing, testing, deploying, hosting, and compressing models.

What is Foundry?

Multiverse Foundry is the workspace behind Multiverse Computing's AI Service and CompactifAI. It's where you bring models in, test them, turn them into live hosted endpoints, and compress them to run cheaper and faster — all from one dashboard at foundry.sandbox.multiverseapp.ai.

How it works​

  1. Get a model into Model HUB — Import a public model from Hugging Face, or use one already in your catalog. See Model HUB and the upload walkthrough.
  2. Test it in the Playground — Send it prompts, compare temperature and max token settings, and watch token count, latency, and cost before you commit to anything. See Playground.
  3. Deploy (host) it — Turn a ready model version into a live, hosted inference deployment with its own CPU/GPU/memory allocation. See Deploy & host a model.
  4. Compress it — Run a CompactifAI job to shrink a model for a target use case (e.g. agentic tool calling) while healing back quality. See Run a compression job.
  5. Monitor everything — Track deployment health and resource usage from Model Deployments, and job status from Compression Jobs.

What's in this guide​

The Product guide section breaks down each screen in Foundry and what every control on it does. The Walkthroughs section gives you copy-paste-able, step-by-step instructions for the four things you'll do most: uploading a model, deploying/hosting it, and compressing it.