Personal assistants are the AI apps your customers ask for help. In the free check, up to 12 of your products become AI agents and read their own pages and each other’s.
It turns the front and the back of your store into AI agents that talk to each other. You only answer the decisions.
24/7/365
What it reads depends on how your store connects. A public website gives products, prices, stock. Orders, carts and customer messages need a deeper connection.
Planned
ExampleA store that also sells online
Examples from MindLoop’s tabs.
Example
Its stock check found your Gift box sold out while shoppers still had it in their carts.
In your Telegram
Gift box just ran out: 0 left. Restock it, or leave it for now?
RestockLeave it
An assistant asks for the whole thing at once. MindLoop turns it into one clear offer your store can keep, and asks you first.
The films are examples.
When an assistant asks, MindLoop answers with your store’s real products, prices and times: a yes your store can keep, a useful alternative or an honest no.
On its own, MindLoop reads, analyses, tests and drafts.
Only decisions cross over.
MINDLOOP13 parts work inside, all day.
See what happens inside
Only decisions reach your phone.
Every part runs today for Client 1. The research part runs with people.
Data sources. Store admin, read only, a copy every 5 minutes.
Your store admin’s 10 pages flow into a twin of your store’s data, one by one, each filling its own row: Stock, Products, Prices, Carts, Orders, Fulfilment, Fulfilment costs, Customers, Recent activity, Wishlist.
Only the newest value of each record. If the twin is out of date, MindLoop waits.
Recorded: Client 1, 30 September 2026. In the copy’s prices: 258 verified prices and 106 price on request.
Completed conversation analyses. Recorded: Client 1, 18 to 22 September 2026. 580 messages analysed. 115 conversations. 139 sources quoted.
Reads 25 messages at a time, from where it stopped.
Each analysis quotes the messages it is based on.
Customer conversations stay closed to personal assistants.
The topic asked about most has the longest bar.
These are counts of what customers asked for, not sales.
Your store’s answers use only recent facts.
Margin is revenue minus costs, divided by revenue.
An offer always stays at or above your minimum margin.
A shopper’s special request becomes a request for you.
If a test ends in a tie, MindLoop checks what customers wrote.
A new plan every 3 hours. Automatic work, grouped as: Working on now, Next, Done today, Waiting on you, Needs your OK or a setting.
Every piece of work gets a score from its worth, its urgency, how well it fits your goals, how overdue it is, how useful its past results were and what is new, set against what it costs.
Worth: set in advance for each kind of work.
Urgency: why it matters now.
Fit: how well it serves your 8 goals.
Overdue: how long since it last ran.
Usefulness: how useful its past results were.
Newness: lower when nothing has changed.
Cost: what the work costs; cheaper work ranks higher.
Example: a stock check is worth more, is more urgent and costs less to run than a price test, so it comes first.
It starts them one at a time. One start every 5 minutes. Up to 24 pieces of work a day.
Work starts only if you allowed it and today’s limit has room.
Switched on for Client 1.
Questions it asked itself.
3 of the 7 questions it can ask itself: What have customers asked for that we don’t sell? Which best sellers could run out soonest? What brings customers back for a second order? Each answered from your own records.
4 a day at most. Not the same one twice in 2 weeks. A refused question waits 24 hours.
Up to 16 pieces of work wait in line.
Example: Allow automatic work in Research? Accept or Decline.
It asks this at most once a month.
If you don’t answer, it stops waiting after 7 days.
In Ask, its latest questions are listed, each with a one-line answer.
Works inside daily limits. Automatic work: Up to 24 pieces a day.
Each day at most: 4 store checks, 1 catalogue stock sweep, 3 tests, 2 research tasks, 4 questions to itself, 1 data sources check.
It plans every 3 hours. Each plan starts the best few pieces of work, inside these limits.
It only starts what it can finish today. Work that would run past today’s limits waits for the next plan.
Your decisions and problems never wait for a limit.
Example. Work: Requests, decisions, and completed work.
Stock check on products in carts. Started by MindLoop. Result saved. Telegram: not sent.
Allow automatic work in Research? Started by MindLoop. Waiting for your answer. Telegram: sent.
Customer request. Started by a customer. Waiting for staff. Telegram: sent.
Stock check on products in carts, started by MindLoop. Result saved. All four steps done: Proposed, Accepted, In progress, Done.
A task shows only the answers it allows: approve, reject, accept, decline, clarify, mark done. Not sent on Telegram: nothing here needs you.
Status words: Waiting for staff, Waiting for your answer, Waiting to be planned, In progress, Accepted, Reported done, Needs attention, Partly done, Result to be checked, Waiting for the customer, Cancelled.
Telegram receipt: Sent on Telegram. Not sent on Telegram: MindLoop keeps updates like this in Work. Not sent on Telegram: today’s limit for automatic updates.
At most one test and one study a plan.
A step can read facts, run a test, draft a study or prepare something for your OK.
Recorded, Client 1, 25 September 2026: current price or 10 percent lower, for one product.
Step 1: one product, two options. Its real price and facts come from your store’s twin.
Step 2: 8 reactions for each option. Written by AI, not by people, and without seeing any score.
Step 3: each reaction is matched by meaning to five sentences: 5, very high, I would definitely buy this; 4, high, I would probably buy this; 3, unsure, I am unsure and could go either way; 2, low, I would probably not buy this; 1, very low, I would definitely not buy this.
Each reaction counts most for its closest sentence. Each level is written 3 ways, and the three results are averaged.
Step 4: the score is how much of the reactions’ weight lands on probably buy or definitely buy: 43.2 percent for the current price, 51.5 percent for 10 percent lower.
The ranges overlap. Forecast, not recorded sales.
Recorded, Client 1, 25 September 2026.
Step 5: the range shows how much the 8 reactions disagree. The more they disagree, the wider the range. A wider range means less sure.
Current price: 43.2, range 34.7 to 51.8. 10 percent lower: 51.5, range 46.2 to 56.9.
Step 6: An option wins only if the lower end of its range is above the upper end of every other range. 46.2 is not above 51.8. The ranges overlap.
What happens next: a clear winner asks you in Telegram. When the ranges overlap, it checks what customers wrote.
16 generated reactions, 8 per option. Forecast, not recorded sales.
Customer Research: prepare, collect, understand. This study is at step 3, understand.
Who answers: people. A study starts only with your OK.
Is a study worth it? It checks: It helps your goals, A decision waits on it, It can change the decision, The right people to ask, Not answered before, How important, How urgent, How useful, It can be done, Easy for the people asked, What it costs. Each with a reason and sources, and 2 to 5 other possible reasons.
Which study comes first: high importance and high urgency come first, low and low come third.
Every finding counts: Could take part, Invited, Said yes, Finished, Stopped, Left out. 6 counts on every finding.
Answers grouped into themes, then findings. What speaks for it, what speaks against it, and what a few said.
Recorded: Client 1, 8 to 9 September 2026. 5 completed interviews, 45 accepted answers, 5 findings, 1 report. The participants were people.
Your rules. What MindLoop may do on its own.
What matters to you: Revenue, Margin, Cash, Waste, Risk, Learning, Customer experience, Compliance. You weigh each one. Set none and all 8 count the same.
14 areas. You switch 13 of them on or off.
What you can allow: Start managed work, Read current business facts, Run a test, Draft a research study, Assess which customer questions need research, Ask assigned staff, Prepare a saved cart for review, Prepare private suggestions for your OK.
Permission comes only from these switches.
On its own: reads, analyses, tests, drafts. Yours to decide: contacting anyone, orders, publishing, price changes.
Hard lines: Payment detail changes, Custom legal terms, Regulated or medical commitments, Private data requests, Unverified suppliers. Always blocked.
It asks before it acts on anything that is yours to decide.
This one reaches you.
Example messages in Telegram. Your weekly review.
Change the price or offer of this product? AI-written reactions, not real customers, preferred 10% lower. Checked before it was sent: in stock this week, above your minimum margin, price may be changed. Accept or Decline. Accept means you change the price yourself. MindLoop then checks your store for it.
Only winning predictions reach you: one option clearly wins, or your customers’ own words settle a tie. MindLoop never changes a price itself.
How it sorts what to send: decision, needs you, result, finding, or kept in Work.
What comes first: Decisions, problems, order receipts, then A teammate’s finished task, customer requests, then Verified results, then Finished tests and research, then Waiting for staff, then Everything else, then Planned. A finding ranks by how important it was rated.
Findings, and questions its work needs answered: at most 3 a day.
Your update level only changes how many findings you get. Decisions and problems always come through.
This one reaches you.
What MindLoop does in your store. Your store’s first offer always comes to you. The steps: special request, then your decision, then one exact offer, then your store’s checkout.
Your store takes the order and the payment.
Example: Checkout is ready. Open checkout.
It checks for the payment after 1 minute, then 2, 4 and so on, up to 60 minutes apart.
Up to 60 minutes apart.
Personal assistants. A customer’s assistant can make up to 1,000 requests, for up to 30 days.
One exact offer. The customer accepts exactly the offer shown. Always at or below your normal price.
Exact offers your store charges: unchanged, then another pack size, then cheaper delivery, then another quantity, then the smallest discount. Only within the limit you set.
Verified results: Outcomes checked against real evidence.
The one verified rule: From your real store, Finished, Checked by MindLoop’s server, Real outcome evidence, Not a test or a generated record, One amount in one currency.
One plain answer. Each finished piece of work comes back as one plain answer, with a reason grounded in evidence: a headline, one action for you, how important it is, and what kind of finding it is.
How important, from 1 to 10: 1 is routine, 5 is worth knowing, 10 is money, customers or stock won or at risk now.
Your weekly review. Never two within 6 days. It tells you what it found, what waits on you, and what it looks at next.
How important its past results were, and your Accepts and Declines, help rank its next plan.
Reads your store
Only the newest value of each record. If the twin is out of date, MindLoop waits.
Only decisions reach your phone.
Every part runs today for Client 1. The research part runs with people.
Data sources. Store admin, read only, a copy every 5 minutes.
Your store admin’s 10 pages flow into a twin of your store’s data, one by one, each filling its own row: Stock, Products, Prices, Carts, Orders, Fulfilment, Fulfilment costs, Customers, Recent activity, Wishlist.
Only the newest value of each record. If the twin is out of date, MindLoop waits.
Recorded: Client 1, 30 September 2026. In the copy’s prices: 258 verified prices and 106 price on request.
Completed conversation analyses. Recorded: Client 1, 18 to 22 September 2026. 580 messages analysed. 115 conversations. 139 sources quoted.
Reads 25 messages at a time, from where it stopped.
Each analysis quotes the messages it is based on.
Customer conversations stay closed to personal assistants.
The topic asked about most has the longest bar.
These are counts of what customers asked for, not sales.
Your store’s answers use only recent facts.
Margin is revenue minus costs, divided by revenue.
An offer always stays at or above your minimum margin.
A shopper’s special request becomes a request for you.
If a test ends in a tie, MindLoop checks what customers wrote.
A new plan every 3 hours. Automatic work, grouped as: Working on now, Next, Done today, Waiting on you, Needs your OK or a setting.
Every piece of work gets a score from its worth, its urgency, how well it fits your goals, how overdue it is, how useful its past results were and what is new, set against what it costs.
Worth: set in advance for each kind of work.
Urgency: why it matters now.
Fit: how well it serves your 8 goals.
Overdue: how long since it last ran.
Usefulness: how useful its past results were.
Newness: lower when nothing has changed.
Cost: what the work costs; cheaper work ranks higher.
Example: a stock check is worth more, is more urgent and costs less to run than a price test, so it comes first.
It starts them one at a time. One start every 5 minutes. Up to 24 pieces of work a day.
Work starts only if you allowed it and today’s limit has room.
Switched on for Client 1.
Questions it asked itself.
3 of the 7 questions it can ask itself: What have customers asked for that we don’t sell? Which best sellers could run out soonest? What brings customers back for a second order? Each answered from your own records.
4 a day at most. Not the same one twice in 2 weeks. A refused question waits 24 hours.
Up to 16 pieces of work wait in line.
Example: Allow automatic work in Research? Accept or Decline.
It asks this at most once a month.
If you don’t answer, it stops waiting after 7 days.
In Ask, its latest questions are listed, each with a one-line answer.
Works inside daily limits. Automatic work: Up to 24 pieces a day.
Each day at most: 4 store checks, 1 catalogue stock sweep, 3 tests, 2 research tasks, 4 questions to itself, 1 data sources check.
It plans every 3 hours. Each plan starts the best few pieces of work, inside these limits.
It only starts what it can finish today. Work that would run past today’s limits waits for the next plan.
Your decisions and problems never wait for a limit.
Example. Work: Requests, decisions, and completed work.
Stock check on products in carts. Started by MindLoop. Result saved. Telegram: not sent.
Allow automatic work in Research? Started by MindLoop. Waiting for your answer. Telegram: sent.
Customer request. Started by a customer. Waiting for staff. Telegram: sent.
Stock check on products in carts, started by MindLoop. Result saved. All four steps done: Proposed, Accepted, In progress, Done.
A task shows only the answers it allows: approve, reject, accept, decline, clarify, mark done. Not sent on Telegram: nothing here needs you.
Status words: Waiting for staff, Waiting for your answer, Waiting to be planned, In progress, Accepted, Reported done, Needs attention, Partly done, Result to be checked, Waiting for the customer, Cancelled.
Telegram receipt: Sent on Telegram. Not sent on Telegram: MindLoop keeps updates like this in Work. Not sent on Telegram: today’s limit for automatic updates.
At most one test and one study a plan.
A step can read facts, run a test, draft a study or prepare something for your OK.
Recorded, Client 1, 25 September 2026: current price or 10 percent lower, for one product.
Step 1: one product, two options. Its real price and facts come from your store’s twin.
Step 2: 8 reactions for each option. Written by AI, not by people, and without seeing any score.
Step 3: each reaction is matched by meaning to five sentences: 5, very high, I would definitely buy this; 4, high, I would probably buy this; 3, unsure, I am unsure and could go either way; 2, low, I would probably not buy this; 1, very low, I would definitely not buy this.
Each reaction counts most for its closest sentence. Each level is written 3 ways, and the three results are averaged.
Step 4: the score is how much of the reactions’ weight lands on probably buy or definitely buy: 43.2 percent for the current price, 51.5 percent for 10 percent lower.
The ranges overlap. Forecast, not recorded sales.
Recorded, Client 1, 25 September 2026.
Step 5: the range shows how much the 8 reactions disagree. The more they disagree, the wider the range. A wider range means less sure.
Current price: 43.2, range 34.7 to 51.8. 10 percent lower: 51.5, range 46.2 to 56.9.
Step 6: An option wins only if the lower end of its range is above the upper end of every other range. 46.2 is not above 51.8. The ranges overlap.
What happens next: a clear winner asks you in Telegram. When the ranges overlap, it checks what customers wrote.
16 generated reactions, 8 per option. Forecast, not recorded sales.
Customer Research: prepare, collect, understand. This study is at step 3, understand.
Who answers: people. A study starts only with your OK.
Is a study worth it? It checks: It helps your goals, A decision waits on it, It can change the decision, The right people to ask, Not answered before, How important, How urgent, How useful, It can be done, Easy for the people asked, What it costs. Each with a reason and sources, and 2 to 5 other possible reasons.
Which study comes first: high importance and high urgency come first, low and low come third.
Every finding counts: Could take part, Invited, Said yes, Finished, Stopped, Left out. 6 counts on every finding.
Answers grouped into themes, then findings. What speaks for it, what speaks against it, and what a few said.
Recorded: Client 1, 8 to 9 September 2026. 5 completed interviews, 45 accepted answers, 5 findings, 1 report. The participants were people.
Your rules. What MindLoop may do on its own.
What matters to you: Revenue, Margin, Cash, Waste, Risk, Learning, Customer experience, Compliance. You weigh each one. Set none and all 8 count the same.
14 areas. You switch 13 of them on or off.
What you can allow: Start managed work, Read current business facts, Run a test, Draft a research study, Assess which customer questions need research, Ask assigned staff, Prepare a saved cart for review, Prepare private suggestions for your OK.
Permission comes only from these switches.
On its own: reads, analyses, tests, drafts. Yours to decide: contacting anyone, orders, publishing, price changes.
Hard lines: Payment detail changes, Custom legal terms, Regulated or medical commitments, Private data requests, Unverified suppliers. Always blocked.
It asks before it acts on anything that is yours to decide.
This one reaches you.
Example messages in Telegram. Your weekly review.
Change the price or offer of this product? AI-written reactions, not real customers, preferred 10% lower. Checked before it was sent: in stock this week, above your minimum margin, price may be changed. Accept or Decline. Accept means you change the price yourself. MindLoop then checks your store for it.
Only winning predictions reach you: one option clearly wins, or your customers’ own words settle a tie. MindLoop never changes a price itself.
How it sorts what to send: decision, needs you, result, finding, or kept in Work.
What comes first: Decisions, problems, order receipts, then A teammate’s finished task, customer requests, then Verified results, then Finished tests and research, then Waiting for staff, then Everything else, then Planned. A finding ranks by how important it was rated.
Findings, and questions its work needs answered: at most 3 a day.
Your update level only changes how many findings you get. Decisions and problems always come through.
This one reaches you.
What MindLoop does in your store. Your store’s first offer always comes to you. The steps: special request, then your decision, then one exact offer, then your store’s checkout.
Your store takes the order and the payment.
Example: Checkout is ready. Open checkout.
It checks for the payment after 1 minute, then 2, 4 and so on, up to 60 minutes apart.
Up to 60 minutes apart.
Personal assistants. A customer’s assistant can make up to 1,000 requests, for up to 30 days.
One exact offer. The customer accepts exactly the offer shown. Always at or below your normal price.
Exact offers your store charges: unchanged, then another pack size, then cheaper delivery, then another quantity, then the smallest discount. Only within the limit you set.
Verified results: Outcomes checked against real evidence.
The one verified rule: From your real store, Finished, Checked by MindLoop’s server, Real outcome evidence, Not a test or a generated record, One amount in one currency.
One plain answer. Each finished piece of work comes back as one plain answer, with a reason grounded in evidence: a headline, one action for you, how important it is, and what kind of finding it is.
How important, from 1 to 10: 1 is routine, 5 is worth knowing, 10 is money, customers or stock won or at risk now.
Your weekly review. Never two within 6 days. It tells you what it found, what waits on you, and what it looks at next.
How important its past results were, and your Accepts and Declines, help rank its next plan.
Only what needs you. Decisions and problems always reach you. You choose the rest.
One tap. Your tap approves that one thing, nothing more.
In Telegram. On your phone, wherever you are.
Your team too. In the film, the kitchen and the floor accept their part the same way.
Table frees at 19:40.
Approve 19:45?19:41
Approved19:41
Booked for 19:45. The kitchen and the floor accepted.19:44
16 generated reactions, 8 per option, for one product.
Forecast, not recorded sales.
MindLoop writes the reactions a customer might have to each option, then scores them. They are written by AI, not by real shoppers.
MindLoop writes the reactions a customer might have to each option, then scores them. They are written by AI, not by real shoppers.
Forecast, not recorded sales. 16 generated reactions, 8 per option, for one product.
One client’s work in September 2026, names left out. It is not a store from the films. The numbers count work done, not sales.
One client’s work in September 2026, names left out. It is not a store from the films. The numbers count work done, not sales.
New terms come to you first. Routine answers stay inside the rules you set.
New terms come to you first. Routine answers stay inside the rules you set.
MindLoop connects your side once. An assistant can only ask for what your store shows and allows.
ONE CONNECTION
WHAT AN ASSISTANT CAN DO
Only with the customer’s OK. They can take it back.
Their assistant: Tonight, 19:30. Window table, daisies, 2 alcohol‑free spritzes.Table at 19:30?19:24
Window table, daisies and spritzes.Ready at 19:45.Dinner until 20:30.Book it?19:43Window table.Ready at 19:45.Book it?19:43
Each assistant connection is checked before it is used.
ONE CONNECTION
Their assistant: Tonight, 19:30. Window table, daisies, 2 alcohol‑free spritzes.Table at 19:30?19:24
Window table, daisies and spritzes.Ready at 19:45.Dinner until 20:30.Book it?19:43Window table.Ready at 19:45.Book it?19:43
Each assistant connection is checked before it is used.
What an assistant can do
Only with the customer’s OK. They can take it back.
MindLoop connects your side once. An assistant can only ask for what your store shows and allows.
Start with the free store check. If you want MindLoop, we set it up with you.
Every connection and permission is agreed with you.
Example: a table tonight.
Example: two bikes and a child seat.
Example: a cake that says OUR PLACE.
Example: a table tonight.
Example: two bikes and a child seat.
Example: a cake that says OUR PLACE.
