Skip to content

Free to use. Demo city Atlanta. Taste by place, never identity.

Put the billboard where the fans live.

Wallspace reads Qloo taste data block by block, picks the billboards, bus shelters and street screens where a brand's audience lives, and writes a line for each face. Atlanta to start.

PlansPorchlight Comedy Tour
Send toExport
#FaceNeighbourhoodTypePer weekWhat the block lovesCopy lineSign-off

sample brand Porchlight Comedy Tour is invented; face rates are sample values. Rows come from the app's public sample plan.

How it works

One brief in. A plan you can defend out.

Four steps, each done by a named model with its cost in the run log. The sample plan below is a real run for a sample brand.

Audience heat · AtlantaQloo
Taste by place

Find the blocks that already love what you sell

Describe the audience by what it loves. Qloo's taste data scores every block in Atlanta against those loves, and each face is scored by the block it stands on, not by who drives past it.

See the heat map
Site photo reads
Site photo read

Drop the faces nobody can see

Nemotron 3.5 Super VL reads the site photo of each shortlisted face for sightline, clutter and how much of the frame the face fills. A board half hidden behind poles leaves the plan before it costs anything.

Read the run log
A face and its lines
A line per block

Write for the block, not the city

Nemotron 3.5 Lightning writes three lines per face from what that block over-indexes on. A live guard checks every line as it lands: plan by taste, never by race, religion, age or any other identity.

Open the picks
Review and sign-off
Review and sign-off

Reviewed once, signed off line by line

Nemotron 3 Ultra 550B reads the finished plan for clashing copy, budget sense and brand fit, and can send a line back. Then NVIDIA NeMo Guardrails, run as a Nebius job, signs off every line before print.

See the sign-off

Everything a buy needs, in one plan

The pieces around the picks: how the money splits, what each model did, what the media owner sees and where the plan goes next.

Budget split with a cap

Spend spreads across neighbourhoods; none takes more than 30%.

See the split
Budget split

Every step on the record

Model, time, tokens and cost for each call.

Open the run log
Run log

The media owner's side

Who lives around each face, by taste, and which brands to pitch.

Open media-owner mode
Audience card

Sent where buyers work

Approval in Slack, the brief to the media owner, a deal, a sheet and the dates.

See the connectors
Slack approval

Taste in. Identity out.

Wallspace plans by what a block loves, aggregated by place. It never sees a person, a phone or a profile, and every number below is measured.

  1. Taste only, never identity

    A brief that targets race, religion, age or any other identity is refused, with a rewrite built on what the audience loves instead.

  2. Every step names its model and cost

    The run log lists each call: which model, how long, how many tokens and what it cost. Nothing hides behind a single "AI" label.

  3. Plans flow into the tools you use

    A Slack approval, an email to the media owner with replies sorted, a HubSpot deal, a Google Sheet, calendar dates and an OpenOOH deal-request file.

    6 destinations

    Slack, Gmail or Outlook, HubSpot, Google Sheets, Google Calendar, OpenOOH

    Source: connections page

From the sample plan

Lines written for the block they stand on.

Each line below is the one the sample plan runs on that face, written by Nemotron 3.5 Lightning and signed off by NeMo Guardrails.

Plans go where buyers already work

Approve in Slack, send the brief to the media owner by Gmail or Outlook and sort the replies, open a HubSpot deal, share a Google Sheet, put the dates on a calendar, or export an OpenOOH deal request.

Connect your tools

Questions, answered plainly

How the taste data works, what it never touches, and what is still a sample.

Where does the taste data come from?
From Qloo's Taste AI, a cultural-affinity engine: what people who love one thing also love, aggregated by place. Wallspace asks it where a brief's loves concentrate across Atlanta and what each face's block over-indexes on compared with the city.
Do you track people or phones?
No. There is no device data, no location panel and no profile of anyone. Every signal is aggregate taste for a place, so Wallspace can say what a block loves but never who walks past a board.
What happens to a brief that targets an identity?
It is refused before planning starts, with a rewrite built on tastes instead. The same taste-only policy checks every copy line as it is written, and NVIDIA NeMo Guardrails signs each line off again before print.
Which models run, and what does a plan cost?
NVIDIA Nemotron 3.5 Super VL plans and reads site photos, Nemotron 3.5 Lightning writes the lines and Nemotron 3 Ultra reviews the plan, all on Nebius Token Factory; Qwen3 Embedding 8B maps free text to Qloo tags. The measured cost per plan is on the limits page, and every step's cost is in the sample run log.
What is real and what is sample?
Real: billboard positions from OpenStreetMap, City of Atlanta neighbourhoods, model runs and their costs, guard results. Sample: the brand Porchlight Comedy Tour, faces added where OpenStreetMap has none, face sizes and rates. Each sample value is labelled where it appears.
Is it free?
Yes. There are no plans, tiers or payments. Wallspace is a hackathon build on Qloo's evaluation key, so it is not offered for production buying.
Which cities does it cover?
Atlanta only for now. The planner, the guard and the connectors are city-agnostic; the inventory and neighbourhood boundaries are not yet.