Built for agentic commerce
Shoppers send agents. Your store answers.
Every live Fluid storefront serves the files and protocols AI agents read: an llms.txt, Markdown and JSON versions of each page, and a Universal Commerce Protocol profile with catalog search, carts, checkout, and orders.
Shoppers are handing purchases to AI.
A shopping agent is an AI assistant a person asks to find, compare, and buy for them. It looks small next to your website today. It’s growing faster than any channel your team has added, and if an agent can’t read and buy from your store, that shopper never sees you.
people used Muse in its first six weeks. Two years ago, AI traffic to retail sites was just starting to climb. Each new wave of agents arrives faster than the last.
Labeled points: Adobe Analytics (Mar 2025), Google (Aug 2026), Meta Q3 2026 earnings call, Poppy 1.0 (Nov 13, 2026). The curve between them is illustrative.of Meta Muse’s weekly users already shop through the agent.
Meta Q3 2026 earnings call, Oct 21, 2026average purchase made through the agent.
Meta Q3 2026 earnings callmonthly users of the Gemini app, passed Aug 11, 2026.
Google, Aug 2026monthly users of AI Mode in Google Search, counted separately.
Alphabet Q2 2026 earnings“Find me the best protein powder.”
The shopper says one sentence. Everything after that happens between their agent and your store, in agreed formats called protocols. Step through it, or let it play.
The shopper asks in one sentence.
Plain languageFind me the best protein powder. Something I’ll actually like.
No search box and no browsing. From here on, the agent does the shopping.
It remembers where they shop.
With permissionWith the shopper’s permission, the agent checks their email and sees they’ve bought from your store before. So it starts with you.
Poppy points it to the right door.
PoppyGETyourstore.com/.well-known/poppy.json200
Points toyourstore.com/.well-known/ucpUCP
Your store’s Poppy file tells personal agents who you are and where your shopping rules live.
It learns your store and searches it.
llms.txt + UCPRead/llms.txt and product .md pages200
UCPcatalog search: "protein powder"3 results
It reads what you sell from llms.txt and each product’s Markdown page, with live prices, then searches your catalog through UCP.
It asks your store a question.
A2AMuseIs the vanilla dairy-free, and when would it arrive?
Your store’s agentAnswers from your product data and shipping rules.
Before it recommends anything, it asks what a shopper would ask a rep. Your store’s agent answers.
It builds the cart and checks in.
UCPMuseFound a vanilla protein from the store you’ve ordered from. Want me to buy it?
When AP2 arrives, the shopper can approve purchases ahead of time, so reorders happen without another tap.
It lands as an order you can trust.
Web Bot AuthThe agent signed its request, so Fluid can check which agent placed the order and mark it verified.
UCP
The rules for shopping.
The Universal Commerce Protocol gives every store the same four steps: find it, put it in the cart, pay, and get the order. Any agent that knows UCP can shop any store that publishes it. It powers buying inside Gemini and AI Mode in Search.
- On FluidEvery live storefront publishes a UCP profile at /.well-known/ucp, with catalog search, Ask, carts, checkout, and orders.
- Paid inside the agentAn agent can pay with Google Pay where your payment provider in Fluid Payments takes it. Otherwise it hands the shopper to your hosted checkout.
A2A
How agents talk to each other.
Agent2Agent lets one agent ask another for help. A shopper’s agent can ask your store’s agent what a shopper would ask a rep: is it vegan, does it ship to Canada, what’s the return window.
- On FluidEach live storefront serves an Agent Card and runs its own store agent. Merchants don’t set anything up.
- Your words onlyThe store’s agent answers from what you’ve published and links back to the page. When there’s no answer, it says so.
Poppy
A signpost for personal agents.
The Personal Agent Protocol, from Meta and Sierra, gives every company one file that tells personal agents like Muse who it is and how to work with it. Muse checks for that file first on every site it visits.
- On FluidEvery live storefront serves /.well-known/poppy.json, pointing agents to its UCP profile, on its Fluid subdomain and its custom domains.
- Two days to shipPoppy 1.0 was released on November 13, 2026. Every Fluid store supported it two days later.
Web Bot Auth
A name badge for agents.
Anyone can claim to be a well-known agent. Web Bot Auth lets an agent sign its request, so a store can check who’s really there instead of guessing.
- On FluidSignatures are checked against the agent’s published key, and each agent order is marked verified, declared, or inferred.
- Numbers you can trustOnly verified orders count as confirmed agent sales, so your agent revenue isn’t padded by guesses.
ACP
Shopping inside ChatGPT.
The Agentic Commerce Protocol does UCP’s job for ChatGPT. It starts with a product feed, so ChatGPT knows what you sell and can send shoppers your way.
- On FluidStores that sell in the US serve an ACP product feed the merchant adds in ChatGPT. Shoppers who find a product there check out on the store.
- Next upFluid’s update for Stripe’s new ACP requirements is coming soon.
AP2
A permission slip for spending.
The Agent Payments Protocol proves the shopper told their agent it could buy this item, at this price, before any money moves. It’s what makes hands-free reorders possible.
- Built on Fluid PaymentsFluid runs its own payments, so AP2 connects to the checkout and processors Fluid Payments already operates.
- Waiting on VisaWe’re waiting on Visa’s guidance. Then AP2 turns on across every storefront. Until then, agents pay through UCP or your hosted checkout.
llms.txt and agents.md
The read-me-first files.
Before an agent can buy, it has to understand the store. These files tell it what you sell, where you sell it, and where everything lives, in a format an AI reads easily.
- On FluidEvery storefront generates llms.txt and agents.md, and a Markdown and JSON version of every product, page, and post, with live prices.
- Your descriptions are the sourceAgents read what your product and page descriptions say, not your design. Clear, complete descriptions are what they repeat.
MCP
A toolbox for your team’s AI.
The Model Context Protocol gives an AI tools it can pick up and use. This one is for your team’s agents, not shoppers’: look up an order, update a product, open a country.
- On FluidThe Fluid MCP server lets Claude, Cursor, VS Code, and other tools work in your company with the access of the person signed in.
Different agents. Different doors.
Each AI assistant reaches a store its own way. Your Fluid storefront answers all of them, and every order lands on the same record.
New protocols ship to your store. You don’t rebuild.
Agent standards move in months. Fluid builds each one into the platform once, so every live storefront gets it on the same day, on its Fluid subdomain and its custom domains.
- On by defaultThere’s no app to install and no migration to schedule.
- Old and new agents both workEach UCP profile serves the 2026-08-25 and 2026-04-08 versions, so agents built on either keep working.
- One recordAgent orders land next to every other order, marked with the agent that placed them.
- Clear statusThe docs label each protocol live or coming soon, so your team knows what an agent can do today.
Questions about agentic commerce.
What is agentic commerce?
Shoppers ask an AI agent to find, compare, and buy products for them. The agent reads a store’s data and buys through shared protocols instead of clicking through its pages.
Which AI agents can shop a Fluid store?
Any agent that reads llms.txt and Markdown pages can learn the store. Agents that speak UCP, such as Gemini and AI Mode in Search, can search, build carts, and check out. Personal agents like Muse find the store through Poppy, and ChatGPT reads the ACP product feed for stores that sell in the US.
Do we have to build anything for agents to read our store?
No. Every live Fluid storefront serves llms.txt, agents.md, Markdown and JSON versions of its pages, a UCP profile, an A2A Agent Card, and a Poppy file.
What’s the difference between UCP and ACP?
Both describe how an agent shops a store. UCP powers buying in Gemini and AI Mode in Search. ACP is the protocol for ChatGPT. Fluid serves both.
How do shoppers buy our products inside Gemini?
Register your store in Google Merchant Center and add a payment account that takes Google Pay. Then agents can pay inside Gemini through UCP.
Can shoppers buy our products in ChatGPT?
Yes. Stores that sell in the US serve an ACP product feed that the merchant adds in ChatGPT, and shoppers check out on the store. Fluid’s update for Stripe’s new ACP requirements is coming soon.
Does Fluid support AP2?
AP2 is built on Fluid Payments, and we’re waiting on Visa’s guidance. Then it turns on across every storefront. Until then, agents pay through UCP with Google Pay or hand the shopper to hosted checkout.
How do we know an order came from an agent?
Fluid records the agent on each order and marks it verified, declared, or inferred. Verified means the agent signed its request with Web Bot Auth and the signature checked out. Only verified orders count as confirmed agent sales.
Do reps get credit for orders an agent places?
Yes. An agent doesn’t click a rep’s link, so Fair Share connects the order to the sponsored customer by their email address and credits that customer’s sponsor. Your Fair Share settings apply to agent orders the same way they apply to every other order.
Does this work on our custom domain?
Yes. Every live storefront serves these files and protocols on its Fluid subdomain and its custom domains.
Can we change what agents read about us?
Yes. Agents read your product and page descriptions, so editing those changes what they see. You can also replace the generated llms.txt with an llms_txt template in your active theme.
What happens when a new protocol comes out?
Fluid builds it into the platform, and it reaches every live storefront without a theme change or an app.
What should we do this week?
Review your top product descriptions and give shipping, returns, and subscriptions their own pages. Then fetch /llms.txt, /.well-known/ucp, and /.well-known/poppy.json on your domain to see what agents get.
Go deeper.
- Every Fluid store is ready for AI shoppersThe release, each protocol, and how to improve what agents read.Read
- Fluid stores are ready for PoppyHow Fluid supported Poppy two days after 1.0.Read
- Every Fluid store speaks A2AYour store gets an agent that answers in your words.Read
- Agentic commerce docsEvery protocol, path, and check, for your developers.Read
Bring us your store. We’ll show you what agents see.
We’ll read your store the way Muse, Gemini, and ChatGPT do, and show you what they find.