It tells you what it found
Analysis, drafts, recommendations. The agent produces a finding, and nothing in the store changes until you decide it should.
Agentic Commerce OS
Influemint is a storefront operated by a workforce of AI agents. They design it, stock it, price it, take the orders and settle the money — and every action that reaches a shopper or a cent stops at your approval first.
Registering provisions the workspace on the spot: a draft store, your agent team, and 100 promotional credits to run them with.
A preview of the Today surface · sample content
The loop
You describe the store. The workforce does the building, the listing, the selling and the accounting — and stops at every door marked with your seal.
Say what you sell and who for. The Assistant plans the build; the storefront and media agents lay out the sections and make the assets.
Catalog and listing agents assemble the products. Pricing computes your share to the basis point and refuses any price that breaks the floor — it never quietly adjusts one.
Publishing waits for your seal. After that, checkout, orders, fulfillment and claims all run on one board, in one status vocabulary.
Earnings post in integer cents against the line snapshot frozen at purchase, and settle to your payout account on a schedule.
Governance
Every action an agent can take is declared in advance as one of exactly three classes. The class follows the consequence, not the agent — and whoever proposes is never the one who approves.
Analysis, drafts, recommendations. The agent produces a finding, and nothing in the store changes until you decide it should.
In scope, undoable, invisible to shoppers — a section redraft, a media variant, a catalog refresh. It applies as soon as it is ready, and Pulse records that it did.
Money, shoppers, live publishing, anything that changes state at a provider. The proposal is prepared in full, then holds — with its consequence spelled out beside the approve button.
One count of what needs you. The same canonical query feeds the rail badge, the Today inbox and every surface that mentions it — so nothing needing your attention is hidden somewhere else.
The workforce
Each agent owns one slice of the store and reports into the same governance plane. They are on demand, not idling: an agent wakes for its assignment, does the work and passivates. The history stays.
The AI control plane
3
Make the store & products
5
Credits · pricing · earnings
3
Orders to fulfillment
4
Reach · insight · domains
3
Sealed — this agent's consequential work always waits for your approval
How it is built
An agent is only as trustworthy as the ledger underneath it. These are properties of the system, not settings you have to go and find.
Integer cents
Money is never a float. Splits compute at basis-point precision, your share rounds down, and the platform absorbs the remainder cent.
Frozen snapshots
A line's price snapshot is taken once, when it is priced, and never recomputed. Earnings and refunds read the same numbers the buyer paid.
Metered before it runs
No model call starts without a credit reservation. What it actually cost is measured and settled afterwards; a failed run refunds the hold.
Refusals are named
When something is not allowed, the system says which rule stopped it — a typed reason, in one sentence. Never a silent clamp, never a blank screen.
Acknowledged is durable
Nothing reports "done" before the write has committed. If you were told it happened, it survived the restart.
Your account is the boundary
Every read, job, event stream and schedule carries your account with it. There is no ambient "current store" anywhere in the system.
In build
These surfaces exist in the workspace and say exactly one thing: planned. You will never find a control here that pretends to do something, or a number that was invented to fill a chart.
Bring a sentence about what you want to sell. The team is already hired.