Three storefronts, three AI assistants —
only one is a platform primitive.
The Fabular reference storefront (“Fresh Haven”) read against two funded Austrian grocers on the only axis that is Fabular’s fight: storefront & AI-commerce mechanics. The obvious read — “Fabular has Jule, it’s ahead” — inverts on inspection.
Two are funded businesses; one is a demo for a platform. So the benchmark splits in two — operational scale (not Fabular’s fight) and storefront & AI mechanics (the whole point). This page is the second layer.
The AI-commerce battleground
Headline · it invertsAll three now ship an assistant, so the naïve scoreboard is wrong. Verified live:
| Capability | Jule · Fresh Haven | Maia · Gurkerl | Alfies AI |
|---|---|---|---|
| Catalog-grounded | ✓ verified (real SKUs + price) | ✓ tool-calls live | gated — unconfirmed |
| Recipes / meal plans | ✓ 3 modes | ✓ + meal plans | “Meine Produkte” |
| Acts on cart / orders | recommends only | ✓ cart · lists · orders | browse |
| Image / photo input | app-only | ✓ on web | — |
| Off-site agentic (MCP) | — | ✓ public MCP | — |
Corrected headline: on AI, Gurkerl is ahead of Fabular, not behind. Maia matches Jule’s grounding and exceeds it on actions and on-web image input. Fabular’s real, defensible edge is narrower: Jule is the only assistant wired natively into a platform-generated storefront — every tenant gets it for free.
Caveat: asked for “vegan dinner for 2 under €15”, Jule led with a €24,99 box — the “budget-optimized” mode is aspirational in the demo, not enforced. Alfies AI sits behind a cookie/address wall (depth unconfirmed).
The decisive mechanic — Gurkerl’s MCP server
Off-site agentic commerceendpoint · https://mcp.gurkerl.at/mcp/ → ChatGPT & Claude Desktop can shop Gurkerl programmatically.
- product.search — by name, category, dietary preference
- cart.* — add / remove / update items
- order.status · history · offers.list · recommendations
A chat widget lives on the site; an MCP server lives where the customer’s agent already is. The store becomes a tool any agent can call — the buyer never visits. Jule has no answer to this channel. But: a grocer builds an MCP server once, as a project. A platform ships it to N tenants as a checkbox. That asymmetry is Fabular’s entire thesis.
↑ Where Fabular’s storefront already leads
- Platform-native conversational commerce — Jule ships with the storefront, no integration
- Producer / regional narrative as merchandising (“Kulinarisches Wien”, per-brand panels)
- One-click recipe→cart bundles · 2-minute onboarding assortment
- Clean Next.js build · DE/EN · trust badges · app Foto-Scan story
↓ Where the incumbents have mechanics Fabular lacks
- MCP / agentic channel (Gurkerl) · assistant that acts + web image upload (Maia)
- Loyalty + paid sub + referral (Gürkchen-Club, Xtra, Empfehlung)
- Food-rescue + discount tier (Rette Lebensmittel, own-brand line)
- Dark mode · live trust telemetry (“12 471+ ratings”) · per-unit / multipack pricing · pure grid mode (Alfies)
Blueprint — what the Fabular platform should ship
Every move = a platform feature- Ship a Fabular MCP server as a platform primitive — one spec, auto-provisioned per tenant (product.search / cart.* / order.status / offers / recommendations). The move only a platform can productize.
- Promote Jule from recommender to agent — add cart/list/order tools; reuse one tool contract behind both the on-site chat and the MCP endpoint.
- Web Foto-Scan + enforce the budget mode — image-to-cart on web, and make “Smarter Einkauf” honor the stated € ceiling (today it doesn’t).
- Loyalty + paid-subscription module — points/club + Xtra-style “free delivery + X % off own-range” + referral.
- Food-rescue / discount-tier merchandising — margin, sustainability story, repeat traffic in one surface.
- Storefront mode switch: narrative ↔ transactional — Fresh Haven is all scroll, Alfies is all grid; let the tenant pick.
- Dark mode theme token · live trust telemetry · per-unit + multipack/carton price-card variant.
Don’t copy: 60-minute q-commerce. Fresh Haven is a direct-from-producer, planned-delivery model — speed isn’t the axis. Agentic reach and platform-productized AI are.
Match the agentic depth, expose the MCP channel, keep the platform asymmetry.
Gurkerl proved the frontier — an assistant that acts, plus an MCP server agents can call. Fabular’s win condition is to ship that frontier as a product every tenant gets for free, not as a feature one merchant built once. Everything else — loyalty, food-rescue, dark mode — is catch-up software a real tenant needs regardless.