AI Search and a Shopping Assistant for Medusa

Medusa gives you a serious commerce backend and a storefront starter you can actually ship.

Interakt gives you semantic AI based search as well as an assistant.

This will give you a search box that understands 'a warm jacket for winter.' Most Medusa stores end up matching substrings against product titles. That works right up until a customer describes what they want in their own words instead of yours.

So we integrated Interakt, an AI-powered search and chat platform, into the Medusa DTC starter. The division of labor is the part worth remembering:

• Medusa owns the product catalog.

• Interakt owns search and chat.

• Products, variants, prices, regions, carts and orders stay exactly where they were.

Note: Medusa recently announced it's own search will be available soon (see announcement), but there's a catch to that, it's only available for running on Medusa Cloud.

Search that ranks, filters, and explains itself

The search page is a normal Next.js page in the storefront, not an embedded widget. Facets are generated from the catalog rather than hand-configured, so brand, category, gender, season, material and color filters appear because the products have those fields, not because someone maintained a list.

Above the results, an AI summary streams in (see the image on top of this article, or the videos linked below to youtube) about a couple of sentences on what was found, plus three suggested follow-up searches. When it fails it renders nothing at all, rather than implying the search itself broke :-)

One detail we're quietly pleased with: The result grid renders through Medusa's own product components, with live region-aware prices. Interakt decides which products and in what order. Medusa stays the source of truth for what they cost and whether you can buy them.

An assistant that actually shops with you

A chat assistant sits on the storefront, knows the catalog, and actually can drive the store. Ask it for a lightweight jacket under a certain price and it searches, talks you through what it found, and takes you to the product page.

Say you'll take the navy one, and it adds that variant to your cart and offers you checkout. Those actions run through the storefront's own Medusa server actions, so the cart it fills is the same cart the rest of the site uses. And you can talk to it. Voice input changes the feel of the thing considerably more than it changes the code.

Discovery is where catalog value often leaks. Every product a customer can't find is inventory you paid for and merchandising you did for nothing. What is cool about this is really that a customer can describe a need in their own words, in their own language, and land on something they'd never have found clicking down a category tree.

None of this changed how Medusa models commerce! And that is how it should be! We hope you agree.

The conversation continues on LinkedIn

If you want to check the conversation and reactions you can also check the related LinkedIn posts:

• I connected Interakt to Medusa

• Here's the plumbing

Want the technical version?

We wrote up the architecture and the decisions we made. That includes the one where we taught the assistant to use the storefront without any client-side tool calling.

You can check that out in the repo if you want. This is inside the Interakt–Medusa Integration (Github) (/2026-09-10-inside-the-interakt-medusa-integration.md).

The implementation is public at alphasolutionsrepo/Interakt-Medusa and Interakt's own Medusa integration guide walks through the setup step by step.