楽AI — Spatial Commerce: test a product against your real room, on your terms.  |  Try the fit demo →
Spatial Commerce • Wave 1

The room that says no.

Scan a room once with the phone in your pocket. Any AI assistant can then measure that room and check whether a product clears the space you actually have — a radius-r clearance sweep plus a physics settle against your real capture, not a picture floating over a camera feed.

Your scan, your room, your key. Nothing renders in your space and nothing is bought without an explicit grant from you.

Try the fit demo Talk to us about a pilot

The clearance, settle, and scale checks are shipped, metered API capability today • checkout is confirm-gated and coming, never autonomous

A simulation of your room, not a picture over it

Every AR shopping app on the market renders a picture of the product in a camera feed. Spatial Commerce tests a product against a simulation of your actual room — the same Gaussian-splat capture, physics-validated geometry, and MCP surface that power Raku Capture, pointed at the fit question.

📏

Clearance, not collage

We sweep the product’s volume against your room’s real geometry and answer whether it clears — a radius-r clearance sweep. It is an honest answer about the space a body this size needs, not a photo trick.

⚖️

Physics settle

The object settles under gravity against the captured surfaces, so “it rests here” is a simulated result, not a guess. The same engine that runs Raku Capture does the work.

📐

Honest about scale

Where a capture’s metric scale is unresolved, the fit answer says so rather than guessing. An absolute measurement is always wrong while scale is unresolved — so we disclose it instead of inventing a number.

What this is not: the shipped fit capability answers clearance and path questions for a swept radius plus a physics settle. It does not yet solve maneuvering an oriented box through a tight corner — the moving-sofa problem — so we never say “the sofa fits.” We say whether a body this size clears the space.

One job, three audiences

A shopper scans a room once. Any AI assistant can measure it, test a product against real geometry and physics, and — only ever behind an explicit human confirm — complete the purchase.

Merchants

Furniture, appliances, home goods — any SKU whose purchase blocker is “will it fit, and how will it look in my space.” Serve a fit answer grounded in the buyer’s real room through your own app or site.

Agentic-checkout rails

A physical-world grounding layer for agent-initiated purchases. The agent about to buy a couch can check it against the hallway first — then hand a confirm-gated cart to a human, never spend alone.

Space owners & shoppers

The whole loop stays on your terms: your scan, your room, your key. Nothing renders in your space and nothing is bought without an explicit, revocable grant.

Five surfaces, ranked by nearness to shipped code

This ordering is the strategy. Each surface is labelled for exactly where it is today — shipped, coming, or a vision we are building the primitives for. Nothing lower on the list is allowed to shape a near-term claim.

A. Try-in-your-real-room

Mostly shipped

Clearance and path checks, physics settle, and honesty-gated metric scale are shipped engine and API capability. The one real build item is SKU→3D ingestion (glTF / USDZ passthrough plus a text/image-to-3D seam), key-gated. The honest claim stays a radius-r clearance sweep until an oriented-bounding-box path-planning lane ships.

B. Agent checkout over MCP

Coming

One confirm-gated checkout tool on the emerging agentic-payment rails (Stripe ACP, Visa TAP, Mastercard Agent Pay). Hard boundary: confirm-then-execute, never autonomous purchase. The human sees the cart and approves; the agent never spends alone. This surface is not live — it ships only after the rails integration and an owner-approved release.

C. Governed spatial placement

Vision • primitive-first

A consent primitive, not an ad product. Deny-by-default blocks brand objects from every captured space; a space owner can grant a scoped, revocable permit; a wearer’s reception key gates what renders on their glasses (dual-key). Placements are physics-real and every render is receipted by signed attestation. There is no ad product, no ad revenue, and no near-term ad revenue line — we build the primitives, honestly labelled, and unbuilt surfaces return an explicit “unavailable” rather than a fabricated impression.

D. Spatial storefronts

Backend shipped • wrapper coming

The content-pack registry and shareable links are merged backend today. The merchant-facing storefront wrapper — branded spatial storefronts on that registry — is the build.

E. Scan-to-shoppable

Vision

A provenance-labelled room inventory, re-pointed from claims to carts: “replace my stuff,” room-to-shopping-list. Every label carries mandatory confidence, source, and observation provenance, and it is gated on real detections and poses — never a fabricated label.

Where this is today

Straight about what is real

Spatial Commerce is a wrapper on the shipped core, not a new engine. Here is the honest split — and it is the split we hold external copy to.

Shipped & metered today

  • Radius-r clearance and path checks against a real capture
  • Physics settle on the captured geometry
  • Honesty-gated metric scale (discloses when scale is unresolved)
  • Content-pack registry and shareable links (backend)
  • The deny-by-default consent spine inherited from the insurance vertical

Not yet — being built

  • Merchant catalog and SKU→3D ingestion
  • Confirm-gated agent checkout (never autonomous)
  • The merchant storefront wrapper
  • Governed-placement enforcement and dual-key consent
  • Any live placement or ad surface

No merchant. No logo. No pilot. No revenue. No merchant has signed, piloted, or paid. Spatial Commerce was promoted to our platform pinwheel ahead of a design-partner commitment, as a disclosed founder call. Every revenue surface above is a ranked thesis, not a booked line — and we would rather tell you that than dress it up.

On your terms

Consent is the spine, not a checkbox

The consent code is inherited from our insurance vertical: deny-by-default, explicit records, honest labelling. Nothing renders in a captured space without an owner grant, and the vision is dual-key — the owner permits placement, the wearer permits reception, both cryptographically attested. Dual-key is a design spec this wave, not shipped enforcement.

Revenue share on governed placement, if it ever launches, is modelled owner-favorable — and only after merchants, supply, and dual-key consent are all real.

Read-path egress

Where your room data goes

RakuAI makes no server-side LLM calls — you bring your own model. A fit query or scene read ships room geometry and semantics to whichever LLM provider you connected. We say that plainly and scope what leaves a private space.

A 3D home capture is inherently identifying, so we never call it “anonymized.” Any data licensing is consent-scoped and provenance-tagged, or it does not happen.

Building a catalog where fit is the blocker?

One design-partner merchant is the conversation we most want to have. If “will it fit” is what stalls your carts, a will-it-fit pilot behind your catalog is where this starts.

Talk to us about a pilot Try the fit demo

Bring your own LLM over MCP • the clearance, settle, and scale checks are live today • checkout and merchant catalog are on the roadmap, honestly gated

The honest fine print

  • Radius-r clearance sweep, not moving-sofa. The shipped fit capability answers clearance and path questions for a swept radius plus a physics settle. It does not solve oriented-box maneuvering through tight corners. We do not imply full furniture path-planning until that lane ships.
  • Checkout never runs alone. Agent checkout is confirm-then-execute behind an explicit human approval, always. Autonomous agent purchasing is a hard product boundary, not a roadmap item — and the checkout surface is not live yet.
  • Placement is primitives-only. Governed placement ships as consent primitives. There is no ad product, no ad revenue, and no near-term ad revenue line. Nothing renders in a captured space without an owner grant.
  • Scan-to-shoppable needs real detections. The labelled inventory produces labels only with real provenance (detections plus poses). No fabricated label may back a shopping list.
  • Scale honesty. Where a capture’s metric scale is unresolved, fit answers disclose it rather than guessing.
  • Bring-your-own-LLM egress. A fit query or scene read ships room geometry and semantics to the LLM provider you connected. We scope what leaves a private space and never label a home capture “anonymized.”