Agents are going to buy things. The only question is who they answer to.
Vendee Labs builds infrastructure for agentic commerce that answers to the buyer.
Our first product, VendeeX, is a shopping agent that runs on your own computer. It cannot be paid to move a merchant up the list, and it keeps a signed record of everything it does in your name — one you own and can read.
Why Now
Three things changed at once.
Agentic commerce is being built this year. The infrastructure decisions being made now will decide, for a long time, whose interests agents serve.
Agents stopped being demos.
People now let software act for them — book, schedule, spend. The question moved from can it to should it, and on whose behalf.
Models got small enough to come home.
Capable models now run on ordinary consumer hardware. Shopping intelligence no longer requires someone else's cloud — which means it no longer requires someone else's business model.
Commerce grew protocols.
Checkout and cart standards are emerging for agents to transact through. The rails are being laid now, and what runs on them is still open.
The Argument
Every agent answers to someone. Most don't answer to you.
An agent that answers only to the buyer cannot be funded by selling the buyer's attention.
Any platform whose revenue depends on placement — ad auctions, sponsored results, commission-weighted ranking — cannot ship a commission-blind agent without dismantling the thing that pays for it.
This is not an engineering gap. It is a structural one. And it is the whole reason Vendee Labs exists.
Pillar One — Local-first
It runs on your computer. That's a position, not a feature.
VendeeX runs on the user's own machine. Identity, memory, cart and purchase record stay there.
We didn't do this because privacy is nice. We did it because a company that monetises attention cannot do it — the moment the agent runs on your hardware and answers to you, the data business behind it stops working.
Local-first is the part of our architecture our competitors would have to break their own revenue to copy.
Pillar Two — Governance
The hard problem isn't capability. It's accountability.
An agent that spends your money needs more than good intentions. It needs to be answerable.
Pre-action checks
Every action is checked against your rules before it happens.
Commission-blind ranking
Enforced in code, not promised in a policy. The ranking engine is never given the number.
Sealed signed record
User-owned, exportable, readable only by you.
Tokenised authority
Counterparties see permission to transact, never identity.
Hard spending limits
Spending limits and approvals the agent cannot override.
Language-safe consent
Safety holds in every supported language — verified unconditionally.
Everyone shipping agents is discovering that the hard problem isn't capability — it's proving what an agent did, and that it was allowed to. We built an answer before the market finished asking the question.
Why It Holds
Motive, mechanism, and law.
Buyer-sovereign
VendeeX has no business reason to collect your data. The revenue model doesn't require it.
Local-first
It mostly can't anyway — your personal records are stored on your device; online features send what an action needs, when you use them.
European privacy standards
Incorporated in Jersey, under a GDPR-mirroring data protection law, in a jurisdiction the European Commission recognises as adequate.
Motive, mechanism, and law — all pointing the same way: at you keeping control.
Three independent reasons, none of which asks you to trust us. A promise only has to fail once. This has to fail three times, in three different systems, one of which is a regulator.
The Primitive
One Avatar. Wherever the agent goes.
The Avatar is the user's identity in the exchange layer — their preferences, their limits, their rules, their record.
It isn't a feature of one app. It's the thing the architecture is built around, and it persists across every release: on the desktop today, and wherever Vendee Labs goes next.
The Architecture
Buying is the first thing we productised. It isn't the only thing the architecture does.
Vendee Labs built exchange primitives into the platform from the start. None of what follows is available today. All of it is what the architecture was designed to carry.
Products
The first product on the layer.
VendeeX
Available NowThe private, local-first shopping agent.
Runs on the user's own computer. Free forever, no account, no email.
The Model
The buyer never pays.
Vendee Labs is targeting two revenue streams: affiliate fees, and a share of PSP transaction fees — the processing fee a merchant already pays, not a charge added to the buyer. Neither one ever touches ranking, discovery, or matching.
Referral rates vary by merchant. That matters, so we'll be plain about it: the incentive to steer exists here exactly as it does anywhere else. The difference is that we made it structurally impossible to act on — no merchant payment reaches the decision layer.
Commission data is withheld in code from discovery, matching and ranking, and confined to a path that only opens after the decision is made.
We also publish a dated public roster of every network and merchant we connect to — because commission-blind ranking is a claim about ordering the candidates, not about which merchants are candidates. Our selection criterion is reach, not rate. You shouldn't have to take that on trust.
About
Who's building this
50+ years in technology · dispute resolution and expert witness on large IT projects · 2nd PhD (Law) on preventing contract failure · AI governance author and adviser
“Governance is established at the identity layer before any transaction, and every action is reconstructable through a signed mandate chain. The file belongs to the buyer, by architecture and not some vague "trust me I'm a mega-tech company of course I have your best interests at heart".”
Our Journey
We built the browser version first.
A hosted buyer agent — conversational discovery across real merchants, multi-vendor carts, multi-item checkout. It worked, and we demoed it publicly.
Then we listened, and we learned.
Real buyers using a real agent taught us things no design document could. What mattered most to them was not what the agent could find. It was what it would do with what it knew about them.
So we took the same architecture and the same principles, and deployed them local-first.
Not a change of direction — a change of ground. The principles didn't move; we found where they go furthest. That's VendeeX.
The beta did its job.
It taught us what to build, and it's earned its retirement. That's what's running now, at vendeex.ai.
Three conversations we're having
Investors
We're building infrastructure for a market that's being wired up right now. If that's your thesis, we should talk.
Merchants
Buyers with intent, arriving through an agent that can't be paid to favour anyone. No advertising spend, no placement auction.
Payment partners
Signed permissions, tokenised authority and an auditable record of every agent action. We create the transaction. Your infrastructure settles it.