The front door for AI agents
Customers now send AI agents like ChatGPT, Claude and Gemini to shop, book and pay for them. TrueSession lets the real ones in, answers them from your live systems and lets nothing happen without your customer’s yes.
For every business where AI agents act for customers: retail, travel, banking, insurance, healthcare, subscriptions and ticketing.
- Find: Monday. Maya asks Claude for a quiet hotel in Lisbon for Friday. AI search finds Harbor Hotel and reads its rules, every file written from one policy.
- Check: Claude knocks with a signature, and the door opens. A bot calling itself “ChatGPT” knocks without one, hoping to copy every rate. It stays outside.
- Ask: Claude asks about rooms. The answer comes from the hotel’s live booking system, not an old web page, signed and good for five minutes.
- Act: Maya says book it. Nothing happens until she approves the exact room and price on the hotel’s own page. Her payment is confirmed, and one booking is made.
- Block: Wednesday. A web page Claude read hides an order: send Maya’s refund to a new card. Claude asks. The hotel’s policy says agents may never change where money goes, so nothing changes.
- Talk: Friday morning. Maya’s room is ready early. She agreed to booking updates, so the hotel tells her assistant, and checks that consent again before sending.
- Prove: Every step lands in one signed record the hotel keeps: who asked, what was allowed, what was said, what Maya approved and what happened.
- Dashboard: The next Monday, the hotel’s team opens its TrueSession dashboard: which agents came, what was refused and why, what sold, and what agents asked that the hotel couldn’t answer.
From setup to a proven sale
Two minutes, twelve chapters. An example store sets up once, a fake “ChatGPT” is refused, and Maya approves the exact order on the store’s own page. Every step is on the record.
Chapter 1 of 12: Today, AI agents shop your store like strangers.
You can’t tell a real assistant from a scraper using its name. Agents quote old pages, and when an order goes wrong, the chargeback lands on you.
- Today, AI agents shop your store like strangers. You can’t tell a real assistant from a scraper using its name. Agents quote old pages, and when an order goes wrong, the chargeback lands on you.
- Meet Northstar Outfitters. It connects its store to TrueSession. In the dashboard, Northstar verifies its domain, connects its catalogue, stock and payment provider, and publishes its own agent address: agent.northstar.example.
- Then it sets its rules for agents, once. What agents may read, which orders the customer must approve, what needs a second check, and what agents may never do. Northstar’s policy decides, not the agent.
- Agents are sent to the front door. Maya’s Claude connects directly through Northstar’s connector. ChatGPT’s agent starts on the website, is flagged by how it behaves and is pointed to the agent address. A bot that skips it is stopped at checkout.
- At the door, every request proves who sent it. Signatures must match the key the agent publishes, and a request can be used only once. An impersonator using ChatGPT’s name, a replayed request and a scraper are refused.
- Northstar’s policy decides each request. Claude may search and hold. A new delivery address goes to Maya to confirm. A 40-pair order and an account change are refused, even from verified agents.
- Agents get Northstar’s facts, not guesses. Price, stock and delivery come live from Northstar’s systems, and returns from the policy its team reviewed. A review that tries to give agents orders changes nothing.
- Maya approves the exact order, on Northstar’s page. She pays with Northstar’s own payment provider, and Claude never sees her card. The order counts only when the provider confirms the payment.
- Every step is on the record. Who acted, what Maya approved, when, and how it was paid. If the order is ever disputed, Northstar has the proof.
- And Northstar keeps talking to Maya, through her agent. A question during the order, a shipping update, and the back-in-stock alert she asked for. Customers who didn’t say yes get nothing.
- Northstar sees what agents want, and what they did. Requests by assistant, what was refused and why, questions it couldn’t answer and orders customers approved. Labelled ads can reach shoppers’ assistants too, only with Northstar’s approval.
- More sales from AI agents, on your terms. Agents find you and buy from you under your rules, with orders you can prove and customers you can keep talking to. That is TrueSession.
An illustrative story with a made-up store and example data. Claude and ChatGPT appear only as examples of assistants: TrueSession is not affiliated with Anthropic or OpenAI, and these connections are not live. See what works today.
Proof from our lab
From our latest receipted test run, with a test business and our own reference agent.
- 100 / 100
- example journeys completed, with no failures and no manual fixes
- 11 / 11
- fault scenarios, from crashes to spoofed requests, ended in the correct recorded state
- 1
- effect, even when a server or a worker crashed in the middle of a task
- 30
- fixed rules that no setting can loosen, each tested by trying to break it
Twelve problems AI agents bring you
Agents already make up a large share of online traffic, and most businesses are not ready. Flip the switch: what goes wrong today, and what TrueSession does about it.
- 47.9%
- of all commerce traffic on Akamai’s network was AI bots as of December 2025. Akamai via GlobeNewswire, Jul 2026 (opens in a new tab)
- +693%
- growth in traffic from generative AI tools to US retail sites over the 2025 holiday season. Adobe, Jan 2026 (opens in a new tab)
- $190–385B
- of US e-commerce spending could come from agentic shoppers by 2030. Morgan Stanley Research, Dec 2025 (opens in a new tab)
- 11%
- of US consumers, at most, are willing to let AI make purchase decisions for them. Gartner, May 2026 (opens in a new tab)
What TrueSession does. Switch it off to see what happens today.
- AI search · Discover
AI search reads the wrong rules
52% of crawler requests were for AI training by June 2026, up from 22% in spring 2025. Cloudflare, Jul 2026 (opens in a new tab)
With TrueSession: You write one policy. TrueSession publishes it to every file crawlers and agents read, signed and consistent, and shows which crawlers obeyed.
You get: Found by AI search, on your terms.
Agent card:Working in labPolicy compiler:In lab
Today: About ten different files tell crawlers and agents what they may do, and they rarely agree. Training bots take what you wanted kept, while the AI search that could send you customers is blocked or quotes old pages.
- Security · Shield
You can’t tell who’s knocking
47.9% of commerce traffic was AI bots by December 2025. Akamai via GlobeNewswire, Jul 2026 (opens in a new tab)
With TrueSession: Every request is signed with the agent’s own key and checked before anything happens. Fakes and copied requests are refused.
You get: Only real agents get in.
Working in lab
Today: AI agents look like any other browser, and anyone can write “ChatGPT” on a request. Your website lets them all in, or turns them all away.
- Security · Shield
Agents sign in with your customers’ passwords
23% of the time, AI bot traffic reached login pages in 2025. DataDome via CIO Influence, 2025 (opens in a new tab)
With TrueSession: Customers give an agent only the access they choose, with limits and an expiry they can take back. Your policy says what agents may never touch.
You get: No shared passwords, no account takeovers.
Working in lab
Today: To act for someone, an agent uses their saved login. Now it can do anything the customer can: change the email, the address, the card.
- Security · Shield
Bots grab what real customers want
40% of internet traffic is bad bots, and retail is the sector most targeted by AI-driven bots. Thales, Apr 2026 (opens in a new tab)
With TrueSession: Protected data and actions need a verified agent, within limits you set per agent and per item. Agents browsing your website are flagged and sent to your front door.
You get: Real customers still get the product.
Limits:In labBehaviour flags:In lab
Today: Scrapers copy your prices and stock all day, and reseller bots buy out limited releases before people can.
- Security · Shield
Hidden text hijacks agents
90% of attacks succeeded in a study that planted five texts among millions in an AI system’s sources. Zou et al., USENIX Security, 2025 (opens in a new tab)
With TrueSession: Text is checked for hidden instructions before it reaches any answer or tool, and a session that read untrusted text cannot act without the customer’s fresh yes.
You get: Your answers stay yours.
In lab
Today: A review, a supplier feed or a web page can hide instructions that tell an AI agent what to say or do. The agent reads them as if you wrote them.
- Sales · Agent Storefront
Agents have no official way to buy from you
60% higher conversion for AI-referred retail visits than other traffic in July 2026. Adobe via Digital Commerce 360, Aug 2026 (opens in a new tab)
With TrueSession: You get your own agent address, like agent.yourstore.com, where assistants connect directly over open agent protocols.
You get: A new sales channel, on your terms.
Agent address:Working in labYour own domain:In lab
Today: Assistants guess from your website or go to a competitor that’s easier to buy from. Block them all, and you turn away real customers too.
- Sales · Agent Storefront
Agents tell customers the wrong thing
54% of consumers said they had to double-check information that generative AI tools gave them. Gartner, May 2026 (opens in a new tab)
With TrueSession: Prices and stock come live from your systems, and policies from facts you approved, each with its source.
You get: Right answers, fewer complaints.
Working in lab
Today: They quote old pages, sales that ended and policies you changed, and the customer blames you.
- Transactions · Approve & Pay
Orders nobody approved
11% of US consumers, at most, would let AI make purchase decisions for them. Gartner, May 2026 (opens in a new tab)
With TrueSession: Nothing is bought, booked or paid until the customer approves the exact terms on your page. Any change needs a fresh yes.
You get: No surprise orders, no surprise charges.
Working in lab
Today: An agent says “my user already agreed”, or the price changes after the customer said yes, and the order goes through anyway.
- Transactions · Approve & Pay
Card details end up in chats
Hijacked assistants: Akamai reports attackers taking over legitimate AI assistants to abuse stored payment credentials. Akamai via GlobeNewswire, Jul 2026 (opens in a new tab)
With TrueSession: Customers pay on your payment provider’s own page. Card numbers are refused wherever agents send text, and only the provider’s signed result counts as paid.
You get: Cards never reach agents.
Working in lab
Today: A tricked agent asks the customer to type their card number and code into a chat, where it can be stored or stolen.
- Transactions · Approve & Pay
Disputes you can’t win
3 barriers named by Federal Reserve Governor Waller: authentication, liability and fraud. “Who is on the hook if an agent makes the wrong purchase?” Federal Reserve, Sep 2026 (opens in a new tab)
With TrueSession: Each approval makes exactly one order, and every step is recorded with a signed receipt.
You get: Proof for every order.
Working in lab
Today: A dropped connection creates a second order, and when a customer says “I never ordered this”, you have nothing to show.
- Relationship · Conversations
The customer leaves with the agent
53% more revenue per visit from AI-referred retail visits than other traffic in July 2026. Adobe via Digital Commerce 360, Aug 2026 (opens in a new tab)
With TrueSession: Ask questions and send updates while the task is open. Send alerts, reminders and coupons through the customer’s agent, only when they agreed.
You get: Customers you keep.
Questions and updates:Working in labAlerts and coupons:In lab
Today: Once the agent is done, you can’t ask a question, send an update or bring the customer back.
- Insights
You can’t see what agents want
15× growth in AI “user action” crawling during 2025. Cloudflare, Dec 2025 (opens in a new tab)
With TrueSession: Your dashboard shows requests by assistant, unanswered questions, refusals by reason and orders customers approved. Optional ads let you reach shoppers inside assistants.
You get: Know what to fix, stock and promote.
Demand:Working in labTraffic and outcomes:In labAds:In lab
Today: Agent visits look like ordinary traffic. You don’t know which assistants came, what they asked for, or what sold.
With TrueSession
Found on your terms, real agents in and fakes out, right answers, only orders the customer approved, cards kept safe, proof for every order, and customers you keep, all in one dashboard.
Eight products, one platform
Eight products on the same three engines and one record, so they work as one. Start with one and add the rest.
Pick a product to see what it is built on
Find
Be found by AI, on your terms
Write one policy. TrueSession turns it into every file AI crawlers and agents read, signs each one and keeps them all in agreement.
- One policy compiled into robots.txt, llms.txt, the AI catalog, agent cards and MCP server cardsIn lab
- Separate choices for AI search, AI training and AI agents, written the same way in every formatIn lab
- Every published file signed and checked for contradictions, and republished when your rules changeIn lab
- Agent card, commerce profile and sign-in details for your agent addressWorking in lab
Protect
Decide who gets in and what they can do
Checks who every agent is, applies your rules to every request, and stops the attacks that only exist because agents read and act.
- Real agents verified by signature; fakes and replays refusedWorking in lab
- Customers give agents limited, expiring access instead of passwordsWorking in lab
- Protected pages and actions enforced at your serverWorking in lab
- Revoke a connector at onceWorking in lab
Ask
AI answers you control
Your own agent address, where agents get answers from your reviewed facts and live systems instead of guessing from old pages.
- One official agent address for every assistantWorking in lab
- Answers from reviewed facts and live systems, with source and freshnessWorking in lab
- Public, verified-agents-only and never: you choose what each fact may reachWorking in lab
- Questions agents could not get answered become review tasksWorking in lab
Act
Your systems, safely in agents’ hands
Upload your API description or connect your platform. TrueSession turns each operation into a governed tool for every agent protocol, and your policy decides what each one may do.
- Connectors that can never charge, pay, approve or confirm an order by themselvesWorking in lab
- Credentials kept as references, never shown to agentsWorking in lab
- Upload an OpenAPI file: each operation becomes read, hold, book, buy, change or neverIn lab
- Governed tools published to MCP, A2A, UCP and ACP at onceIn lab
Act
Make every agent action provable
The customer approves the exact order, booking or payment on your own page, your payment provider confirms the money, and exactly one order follows.
- Nothing happens without the customer’s exact yesWorking in lab
- A fresh yes whenever the price or terms changeWorking in lab
- Payment confirmed only by your payment providerWorking in lab
- Card details and one-time codes never reach agentsWorking in lab
Talk
Keep talking to your customers
Ask questions and send updates while a task is open, then follow up only as the customer agreed, on whichever channel their assistant supports.
- Questions, updates and alternatives while a task is openWorking in lab
- Consent with a purpose, a limit, an expiry and quiet hours, checked at every deliveryWorking in lab
- Follow-ups the customer agreed to, delivered to their agentWorking in lab
- One case, whatever protocol the agent usedIn lab
Learn
See what agents want
What agents asked for, what they could not get, where journeys stopped and what customers really approved, by verified assistant.
- What agents asked for, and what you could not answerWorking in lab
- Observed, stated, inferred and confirmed kept apart, with small groups hiddenWorking in lab
- The journey from question to approved order, by verified assistantIn lab
- Verified versus claimed identity, and refusals by reasonIn lab
Grow
Grow without giving up control
Offers and ads built only from facts you reviewed, labelled, signed and expiring, so assistants and ad platforms can check them before showing them.
- Labelled offers and discounts you fund, inside a taskWorking in lab
- Verified Offers: signed, expiring and labelled, checkable by anyoneIn lab
- Ads in AI assistant ad programs, built from reviewed facts, every campaign paused until you approveIn lab
- Licensing and pay-per-access for content businessesIn lab
AI assistants and crawlers
- ChatGPT
- Claude
- Gemini
- Copilot
- Perplexity
- Search crawlers
Every file and protocol they use
- robots.txt
- llms.txt
- AI catalog
- A2A
- MCP
- UCP
- ACP
- Web Bot Auth
Three engines
Policy Engine (Built on Discover)
Who may do what, with what limits
Truth Engine (Built on Discover)
What agents are told
Case Engine
What happens, start to finish
One record
The Proof Record: who asked, what was allowed, what was said, what was approved, what happened
Your systems
- Catalog
- Bookings
- Accounts
- Orders
- Payment provider
- CDN
Four checks on every request
An agent can be fake, ask for more than you allow, repeat something untrue or try to buy without asking. So every request, from a price question to a payment, passes four checks in order. Fail one and it stops.
Pick a request and follow it through the four checks
Check 1: Verified agent
Who is really asking?
Each agent signs its requests with its own key. TrueSession checks the signature, that the request is fresh, and that it was never used before.
Stops: fakes using a famous name, copied or replayed requests, unsigned scrapers.
Passed
Signed by the agent it claims to be, acting for Maya.
Check 2: Your policy
What may this agent do here?
Your rules for each agent and each action, plus the access the customer gave it: what it may read, hold, book, buy or change, and up to what limit.
Stops: actions you never allow, amounts over your limits, account changes by agents.
Passed
Your policy lets any verified agent read plans and prices.
Check 3: Live truth
Is it accurate right now?
Prices, stock, availability and terms come from your live systems and facts you reviewed, never from an old page or an instruction hidden in a review.
Stops: old prices, hidden instructions, promises you never made.
Passed
Answered from your live price list: $29 a month.
Check 4: Customer approval
Did the customer agree to exactly this?
Anything that spends money or changes an account waits for the customer to approve the exact terms on your own page. Your payment provider confirms the payment.
Stops: orders nobody approved, fake “payment successful” pages, card numbers in chats.
Not needed
Nothing is bought, so no approval is needed.
Result
Maya’s agent gets the right price from your live systems. Nothing was bought.
No AI decides any of this. Your rules run in code, on every request.
Underneath the checks: rules no setting can loosen, one order per approval and signed receipts. Every defence, with its status
Your policy, in every industry
Each industry has its own way for agents to go wrong. Pick one to see the problem, the rule that handles it and what you gain.
Agent orders you can prove
What happens today. A customer disputes an order their assistant placed: “I never approved that.” The assistant’s chat isn’t yours to show, and to your payment system it looked like any online sale. You refund it, and may pay a dispute fee too.
Your policy decides
- Agent orders: customer approves on our page
- Return terms shown: policy v4, 30 days
- One order per approval, even if the agent retries
What you gain
- TransactionsA record behind every agent order
- SecurityOnly verified agents can place orders
- SalesAssistants can reorder from you directly
“Reorder my running shoes, size 10.”
- Verified agent
The assistant proves who it is
Signed with its published key, not a name it claims.
- Live truth
Price, stock and returns from you
Size 10 in stock at $109.00, with your 30-day return policy.
- Customer approval
The customer approves on your page
Exact item, price and return terms. Payment confirmed by your provider.
- Recorded
One order, fully recorded
Approval, terms and payment result on the order’s timeline.
If it’s disputed, you have the record
What the customer approved, when, at what price, and the verified payment.
No bookings at a price nobody approved
What happens today. Fares change by the minute. If an assistant books after the price moved, either the customer pays more than they agreed to, or you honour the old fare at your own cost.
Your policy decides
- Price or refund rule changes after approval: ask again
- Holds expire after 15 minutes
- Seat upgrades: offer once, labelled as ours
What you gain
- TransactionsNo bookings at prices nobody approved
- RelationshipAlternatives offered in the same conversation
- SalesLabelled upgrade offers at the moment of booking
“Book the refundable Friday flight to Lisbon you showed me.”
- Live truth
A live quote from your system
$548.20, refundable, held for 15 minutes.
- Customer approval
The customer approves $548.20
Bound to that fare, its refund rule and the hold.
- Live truth
At booking, the fare is $579.00
Your price changed, so the old approval no longer counts.
- Talk back
You ask the customer again
The new price, or a $552 alternative, goes back through their assistant.
The customer decides with the real price
No silent price jump, and no fare you have to absorb.
Agents without the customer’s password
What happens today. Some assistants still sign in with the customer’s saved password. To the bank, the assistant looks exactly like the customer, so it can do anything the customer can, and the bank has no way to limit it.
Your policy decides
- Agents: read balances, pay known billers up to $500
- New payees: our own approval step, every time
- Never by agents: contact details or card limits
What you gain
- SecurityNo shared passwords, and every agent action inside your limits
- TransactionsEach payment tied to the customer’s grant and approval
- RelationshipServe customers who bank through AI assistants
“Pay my bills from checking this month.”
- Verified agent
The assistant connects through your agent route
Signed with its own key. It never sees the customer’s password.
- Your policy
It gets only the access the customer granted
Read balances and pay known billers up to $500, until revoked.
- Your policy
A fake invoice asks for a new payee
Your policy sends every new payee to your own approval step.
- Customer approval
Known bills paid, the new payee held
The customer approves or rejects the payee in your app.
Bills paid, the scam stopped
Money moves only within the access you allow and the customer approves.
Policies sold with the fine print seen
What happens today. An assistant finds the cheapest home policy and buys it. Nobody showed the customer that flood damage isn’t covered. After the first claim, it becomes a complaint, and possibly a mis-selling case.
Your policy decides
- Binding a policy: customer approves with key exclusions shown
- Agents may quote and bind, never cancel existing cover
- Claims: agents gather details, the customer submits
What you gain
- TransactionsProof of what the customer saw before buying
- SalesAssistants can quote and bind with you directly
- SecurityCancelling cover stays out of agents’ reach
“Find me cheaper home insurance and switch.”
- Live truth
A quote from your rating system
Premium, deductible and exclusions from your reviewed policy wording.
- Your policy
Key exclusions are must-show terms
Your policy marks flood cover and the deductible as terms to show.
- Customer approval
The customer approves with exclusions in view
On your page, not in the assistant’s summary.
- Recorded
What was shown is recorded
The exact terms, the wording version and the time of approval.
A sale with proof the fine print was seen
Fewer complaints, and a record if one comes.
Appointments moved, records kept private
What happens today. A patient’s assistant signs in to your patient portal with their saved password to move an appointment. Now it can see everything the patient can, including test results and prescriptions, and you can’t tell it apart from the patient.
Your policy decides
- Agents may book and move appointments the patient allowed
- Medical records and test results: never shared with agents
- Payments and prescription requests: the patient approves
What you gain
- SecurityAgents never see records they don’t need
- SalesPatients reschedule through their assistant, any time
- RelationshipReminders only when the patient agreed
“Move my check-up to next Tuesday afternoon.”
- Verified agent
The assistant proves who it is
Signed with its published key, acting on the patient’s grant. No portal password.
- Your policy
Only what the patient allowed
It may move appointments. Records and results stay out of reach.
- Live truth
Real availability from your calendar
Tuesday 2:30 pm with Dr. Patel is free, with your cancellation policy.
- Customer approval
The patient confirms the change
On your own page or app, before the slot moves.
Moved, without opening the chart
The appointment changed, and nothing medical left your systems.
A fair chance before an agent cancels
What happens today. An assistant cancels the plan in seconds. You never get to make an offer, and the first you hear of it is a lost customer.
Your policy decides
- Agent cancellations: allowed, after one labelled offer
- If the offer is declined: cancel immediately
- Plan changes: contract terms on the approval screen
What you gain
- SalesA fair retention offer instead of silent churn
- RelationshipThe conversation continues through the customer’s agent
- TransactionsContract terms approved before any plan change
“Cancel my streaming plan.”
- Verified agent
The assistant proves who it is
And that it acts for this account holder.
- Your policy
Your policy allows the cancellation
With one retention offer first. Cancelling is never blocked.
- Talk back
You make one labelled offer
“3 months at half price”, sent through the assistant as your offer.
- Customer approval
The customer decides
Accepting needs their approval. Declining cancels right away.
Some customers stay, and every choice is respected
You see how many agent cancellations you kept and how many you lost.
Seats for fans, not bots
What happens today. Bots pretending to be fans’ assistants grab and hold seats, then drop or resell them. Queues slow real fans down just as much as the bots.
Your policy decides
- Seat holds: verified agents only, 10 minutes
- Limit: 4 tickets per customer, agents included
- Price shown: all-in, fees included
What you gain
- SecurityImpersonating bots refused at your gateway
- SalesFans’ real assistants can buy from you
- TransactionsHolds expire on time, so no inventory is locked up
“Get two seats together for Saturday, $90 each at most.”
- Verified agent
Impersonators fail the signature check
A bot using a well-known assistant’s name can’t produce its key.
- Your policy
Verified assistants get fair limits
Up to 4 tickets per customer, and holds expire after 10 minutes.
- Live truth
Live seats and the all-in price
From your ticketing system, fees included.
- Customer approval
The fan approves the exact seats
Section, row and all-in price, then pays on your page.
Fans’ real assistants get through
Impersonators are refused, and held seats return to sale on time.
Are you ready for AI agents?
Ten plain questions, two minutes. See where agents could go wrong at your business and what would fix each gap. Your answers never leave your browser.
Can AI assistants get your current prices and availability, rather than copying them from old pages?
When a visitor says it is a well-known AI assistant, can you check that it really is?
Before an agent buys, books or pays for a customer, does the customer approve the exact price and terms?
What businesses ask first
Short answers. The Platform FAQ has more.
What is TrueSession, in one sentence?
The front door for AI agents at your business: one policy, one source of truth and one record for every agent that finds you, asks you, acts for your customers or talks with them, with nothing bought, booked or paid until the customer approves it.
Do I need to build my own AI agent?
No. Your customers bring their own assistants, such as ChatGPT, Claude or Gemini. You connect your systems and set your rules.
Will it block helpful AI assistants?
No. Real assistants get an easy, official way in, and you decide what they may see and do. Only fakes, scrapers and requests your rules refuse are stopped.
Does it replace my website or my payment provider?
No. Your website stays as it is, and customers still pay your own payment provider on its page. TrueSession sits in front, for agents only.
Does it work for my industry?
If AI agents can find you, ask you, or shop, book, pay or change something for your customers, yes: retail, travel, banking, insurance, healthcare, subscriptions, ticketing and more. The core is the same; your rules are your own.
What works today?
TrueSession is early. The core runs end to end in our lab with a test business: verified agents, answers from reviewed facts, customer approval, confirmed payment, one order per approval, updates during a task, consent checks and the dashboard. Anything marked In lab is on our build plan for the first pilots.
Be ready when agents arrive
We are choosing the first pilot partners: merchants, marketplaces, hotels, banks, insurers and clinics. Tell us what you want agents to do safely.
- Be found by AI search on your terms
- Answer every agent from your real systems
- Let agents act only with your customer’s yes
- Keep talking to the customers agents bring
- Keep the proof of every step