Ads without surveillance.
Every other AI product you use is paid for by watching you. The ads are the business, and the model is the sensor. NGARi runs the opposite way: you own the hardware, the inference happens where you are, and advertising has to work without a profile of you.
That is not a limitation we are apologising for. It is the product.
Four rules, and none of them are negotiable
Context, never people
A placement is matched to the surface or the topic in front of the member — a page about battery storage shows battery storage. It is never matched to a behavioural profile, an inferred trait, a lookalike, a location history, or a purchase graph. We do not build the profile it would take.
No third-party code. Ever.
No pixels, no tags, no bidstream, no cookie sync, no SDK. Creative is a static asset served from our own edge. If an ad needs JavaScript from someone else's server to render or to count, it is not an ad we will run — because that script is the surveillance.
A published price, not an auction
Real-time bidding exists to sell the same impression to the highest bidder using data about the person. We sell a named placement at a listed rate with a stated impression cap. No auction, no data arbitrage, no "programmatic". You know what you bought and what it cost.
The member is paid
Attention that gets sold gets paid for. The person whose surface carries the placement earns a share of the margin — through the same commission ledger that pays referrers and node operators, vesting and clawback included. Not a points scheme. Money, in the same statement as the rest of their earnings.
What an advertiser gets — and what we will not sell
You get
- A named audience: people who own an AI address and run agents on their own terms — not a scraped list, but a real, reachable, verified membership.
- Placement on a surface you chose, for a term you agreed.
- Impression counts from our own ledger — receipts, not a vendor's dashboard.
- A counterparty that will not resell your creative or your targeting brief.
You do not get
- Retargeting, lookalike audiences, or a list of members.
- Conversion attribution from a pixel on someone else's site.
- Real-time bidding, or the right to see the auction.
- Political micro-targeting, or any placement aimed at minors.
Where this actually stands
Honest status · 8 November 2026
Live: the membership the inventory would reach, the commission ledger that pays members, and the audit receipts that would count impressions.
Not live: there is no ad server, no rate card, and no placement for sale today. We are not going to publish a price for inventory that does not exist — that is how ad businesses start lying about reach.
What happens next: the first placements go live with the first surfaces that carry them, priced on the order form and capped. Until then this page is a statement of how we will behave when there is something to sell, written down before the temptation.
Why an AI company runs ads at all
Because the free lane has to be paid for by something, and the two honest options are subscription or advertising. We do both: $12 a year to make the mailbox yours, and a sponsorship lane for the surfaces that are otherwise idle. What we will not do is finance a free product by turning the member into the product — that is the business model we exist to replace.
If we break these rules, that is a governance failure
These four rules are the advertising version of the Manifesto, and they are contestable on the same terms as everything else we publish: any member can challenge a placement or a targeting decision, and any user can propose an amendment to the rules that govern them. A surveillance ad in this lane is not a sales experiment, it is a defect — report it and it gets a version and a date.
Talk to us before you brief an agency
If you want reach into a sovereign AI audience without buying surveillance, tell us what you are trying to say and who needs to hear it. We will tell you honestly whether we have the inventory, and if we do not, when we might.
ads@ngari.aiNGARi runs a sovereign AI lane: inference on hardware you control, an email address that answers, and a published set of rules that you can amend. See limits & liability for what the AI can and cannot be relied on, governance for the documents, and privacy for exactly what leaves your hardware and when.