Milly Chat in May 2026: Live on the Shopify App Store, Curated AI Output, and Auto-Apply Discounts

May 2026 is the month Milly Chat went from "live on a handful of stores" to publicly installable by anyone on Shopify. The headline is the App Store listing — install in a few clicks, 7-day free trial on every plan. But the same month shipped two new levers for shaping what the AI says (Curated Collection and Curated Responses), a marketing-owned Brand Voice layer, auto-apply discounts at the cart, and a dedicated home for the corrections you teach the AI.

Most of that feature set also graduated from early access to every plan at the same time. For full details on any individual feature, see its dedicated post on /writing.

Live on the Shopify App Store

Milly Chat is now publicly listed on the Shopify App Store. There's no code and no theme edits involved: the widget auto-injects through a Theme App Extension, so installing is a few clicks plus one toggle in the theme editor. Every plan comes with a 7-day free trial.

• Three self-serve tiers. Starter ($39), Essentials ($99), and Core ($599). You can upgrade or downgrade between them in-app from the Billing page — no reinstall, no email to us.
• Zero-touch install. The subscription approval fires on the OAuth callback, so you land on Shopify's approval screen right after install instead of hunting for a billing tab. Approve it and the widget activates on its own.
• Custom distribution still exists. High-volume and headless stores can keep using the custom-distribution path with out-of-band billing. The public listing doesn't replace that; it adds a self-serve option next to it.

From early access to everyone

Earlier in May we added release tracks: a per-store choice between "Stable" and "Latest." New features ship to Latest first, soak against real traffic, then graduate to Stable once they've earned it. It let us keep moving fast without putting every fresh change in front of every merchant on day one.

With the public launch, the soaked May feature set graduated from Latest to Stable — so every merchant on every plan now gets it by default: Brand Voice and Brand Dictionary, Curated Collection, Curated Responses, Quizzes, and the post-purchase analytics. The release-track mechanism stays in place for genuinely new features and partner betas. The one feature still riding Latest is auto-apply discount, which we're keeping in early access a little longer while we tune it (more on that below).

Curated Collection and Curated Responses

Two levers for shaping AI output, paired as a tier progression — scope on Essentials, precision on Core.

• Curated Collection (Essentials+) restricts product recommendations to a single Shopify collection of your choosing, and that scope holds across all four widget formats — search, in-chat recommendations, and the sticky add-to-cart bar. It auto-resyncs whenever the underlying collection or its products change, so there 's no manual "Sync Now" after you reorganize the catalog.
• Curated Responses (Core+) lets you pre-author the exact answer to a known ask, tied to a quick question or trigger. The curated response can take over the turn without falling through to product search, and it injects guidance at the system-prompt level so you control the framing and the must-include details.

The two are covered in depth in their own post: Curated Collection and Curated Responses.

Brand Voice and Brand Dictionary

Two new fields sit next to AI Instructions on Configure → Response, and they exist to split a concern that used to be tangled together.

• Brand Voice describes the persona — "friendly and outdoorsy," "expert and concise" — and it gets layered into every response.
• Brand Dictionary lists preferred terminology and substitutions the AI honors verbatim. "Call them anglers, never fishermen" is the kind of rule that lives here.

The point of the split: AI Instructions stay developer-owned safety rails, while Voice and Dictionary are creative-owned identity that a marketing team can edit on its own without touching the rails underneath. There's a deeper write-up in its own post: Anglers, Not Fishermen.

Auto-apply discount at the cart

When the AI issues a coupon after an email capture, an opt-in toggle now pushes the code straight into the shopper's cart instead of asking them to copy and paste it at checkout.

• Native storefronts — the widget silently sets the discount cookie in the background and the shopper stays on the page they're on.
• Headless storefronts — the widget dispatches a milly:applyDiscount window event for your bridge to consume, a sibling to the existing milly:addToCart callback.
• The code is still in the reply as a fallback ("I've applied a discount to your cart — it'll appear at checkout"), so a shopper can verify it or apply it by hand if anything goes sideways.

It's default off, so existing "use code X at checkout" flows are unchanged for stores that haven't opted in. Available on Essentials and up, and — as noted above — still on the Latest track while we soak it.

A home for corrections

When you correct the AI — a misclassification, a stale product detail — those corrections now have a dedicated lander that summarizes every one of them in a single view. A KPI strip at the top breaks them down by applied, pending, and issue type, and a group-by-issue-type view makes recurring patterns easy to spot. It defaults to the Applied tab so you scan what's already live before wading through pending feedback, with a tier counter against your limit (Starter 3, Essentials 10, Core 50, Enterprise unlimited).

Smaller things, in one paragraph

The Coupon Reward setup gained an autocomplete picker for the Shopify discount template, so you select it by name instead of pasting a bare GID. Empty-state widget responses got smarter across the board: "show me your best sellers" falls back to the newest products when a fresh install has no order history, a fully sold-out catalog returns a clear "none in stock at the moment" message, and password-protected development stores get a dev-store-aware reply — all three guards run before the AI, so cold-start widgets stay responsive instead of hallucinating. Inline action banners across the dashboard were replaced with a consistent toast pattern, so save, sync, and delete confirmations feel uniform everywhere. And under the hood, system-prompt construction moved into a shared builder that both response paths read from, which also lets the response cache bust the moment you change a store's name, plan, or feature flags instead of lagging behind.

What's coming next

A few things queued for June and beyond:

• Reviews integrations — Yotpo first, so the AI can ground product recommendations in real customer feedback alongside the catalog and your knowledge base.
• Cross-sell recommendations — complementary products surfaced after add-to-cart, pairing naturally with the sticky add-to-cart work from April.
• Live-agent handoff — routing a conversation to a human through Zendesk when the AI hits the edge of what it should answer. We greenlit this with a merchant who asked for it directly.
• Per-campaign coupon routing — a direct extension of this month's auto-apply discount: route to different discount codes based on what the AI classified the lead as, or the campaign they arrived from, with support for evergreen automatic discounts.
• Leads insights — capture-rate trends, intent breakdown over time, and a view of which corrections have been applied so you can see how the AI is being tuned.

The App Store launch doesn't change how we build. Merchant requests still drive most of the roadmap — the listing just means more of you can get started without talking to us first.

If something on your store would benefit from a feature we haven't built yet, say so. The easiest way is to drop a note via the contact page, or reply to whatever thread you're already on with us. We read everything, and the request list is where most of the next month's roadmap comes from.