Skip to the page

What Munshi does for your shop, one job at a time.

Each job below is shown working, on a sample shop. Read the ones you care about, or try them yourself in the demo.

Demo on synthetic data · suppliers simulated · payments recorded, never processed

An empty kirana in morning light: full shelves of packets and jars behind a steel counter, a hanging scale.
Before opening
The counter at rush hour: the shopkeeper hands a bag across, phone in his other hand, customers crowding in.
At the counter
At night under a street lamp, the shopkeeper pulls down his shop's shutter, smiling, a dog asleep nearby.
After closing

Morning brief

Know what changed before you lift the shutter.

Overnight, the agents read your books. By morning you have a short list of what changed, with the most money at stake first.

Tap any figure and Why this number walks it back: the read, the book row, the decision, and the person who approved it. A link that cannot be walked says Cannot verify where it breaks.

Before sunrise, the shopkeeper stands in his lit shop reading his phone while the street is still blue and quiet.

This morning

  1. Stock short on the shelf₹2,640
  2. Credit overdue, three customers₹2,350
  3. UPI settled short₹412

Why this number

  1. The figure₹2,640
  2. The readStock against the shelf count · 6:00 am
  3. The book rowStock count · line 7
  4. The decisionCount approved · Tue 9:14 am
  5. The approverThe owner

Example · synthetic data

Shelf count from a photo

Count a shelf by taking its picture.

Point your phone at a shelf. Munshi tells you how many packets are really there and how far that is from what the books say.

No model does the sum. One model names the products, another finds each packet, and plain code counts the boxes and compares them with the books. A gap becomes a finding, and a reorder waits for your yes.

A shelf with stacks of red and yellow packets at either end and a long empty gap between them.Biscuit packet
  1. Muse Sparknames the product
  2. SAM 3.1finds each packet
  3. Plain codecounts them
Books say
34
Shelf shows
21
Short by
13

Example · synthetic data

Supplier bills

Photograph a bill. Skip the typing.

A crumpled paper bill or a PDF becomes a purchase draft, line by line, ready for you to check against the goods.

The figures are kept as cells of the stored bill. The line text is shown as it was read, with a note that it is not checked. Stock and the books move only after you approve.

Over the shopkeeper's shoulder: his hand flattens a crumpled supplier's bill while his phone photographs it.

Purchase draft

Sample Traders
  • Sugar50 kg₹2,150
  • Mustard oil12 L₹1,860
  • Tea dust8 kg₹1,440
  • Total₹5,450

Item names as read, not checked · From a simulated supplier

Example · synthetic data

Ask

Ask your books out loud, in Hindi or English.

Speak or type the way you talk at the counter. The answer comes from your books, and every figure in it opens its row.

The model writes the words; the numbers come from stored rows. An answer that cannot be checked is not shown at all. You see Cannot verify, and the reason.

A shopkeeper in a checked shirt speaks into his phone at the counter, jars and packets behind him.

Aaj kitna udhaar wapas aaya?

How much credit came back today?Asked by voice

Munshi

₹750 came back today, from Sunita D. Three customers still owe ₹2,350.

Bank mein kitna hai?

Cannot verifythe bank statement has not been imported

Example · synthetic data

Approvals and the stamp

Your yes is the only thing that changes the books.

Agents write fixes as drafts: a reorder, a reminder, a stock correction. Each one waits in your tray, with exactly what would change.

When you approve, Munshi first reads every cited number again. If one has moved, the draft goes stale instead of running. If not, it runs under your name, with its audit entry, in one step.

Your tray

Drafted by DashReorder · red biscuit packets13 packets at ₹20 · ₹260Waiting for youApprovedby the owner

Ran under the owner's name, with its audit entry.

Example · synthetic data

Credit and PayPal links

Get your udhaar back without the awkward call.

The credit book shows who owes what, and since when. For an overdue balance, Munshi drafts a polite reminder and can add a PayPal link.

The link and the message are separate drafts, each approved by you. When a paid notice comes in, it does not touch the books: it becomes a finding, and its fix is a draft to record the payment. Munshi never moves money.

The shopkeeper writes in the credit book at the counter and looks up as a customer waves from the bright street.

Credit book

  • Rakesh K.41 days₹1,200
  • Anil P.36 days₹750
  • Mohan L.31 days₹400
Reminder · draft

Namaste Rakesh ji, aapka ₹1,200 ka hisaab 41 din se baaki hai.

  1. Paid notice
  2. a finding
  3. a draft to record it

Example · synthetic data

The agents' office

See who worked on what, and read their notes.

Seven agents, each with a desk: Munshi leads, and the others look after stock, money, the counter, customers, campaigns and memory.

Nothing they do is hidden. The office shows the last seven days of runs, what each one proposed, and where it stopped. Open a run to read it step by step.

  • Munshithe lead
  • Tallystock
  • Ledgermoney
  • Dashcounter and fixes
  • Echocustomers
  • Bloomcampaigns
  • Keepermemory

Run log

Tally · shelf gap · 6:02 am

Steps of this run

  1. Read the booksstock against the count42 ms
  2. Model callsam-3.11.8 s
  3. Checkedevery figure has its cell6 ms
  4. Proposeda reorder, waiting for the owner11 ms

Example · synthetic data

Campaigns and channels

Bring regulars back with a message you approved.

Pick who to reach, on WhatsApp or Instagram. Bloom writes a first version; you edit it, send yourself a test, and only then approve.

Replies land in one inbox, where Echo drafts answers and orders for you to approve. An approval covers the exact picture: change the picture and the draft goes stale. On this demo, no message leaves the shop.

An empty kirana in morning light: full shelves of packets and jars behind a steel counter, a hanging scale.

Diwali week: fresh stock is in, and your usual is set aside. Come by this evening.

WhatsAppInstagram

  1. Bloom's draft
  2. Test sent to you
  3. Waiting for you

Your tray

Your storefront

A shop site that takes bookings while you work.

A shop can have its own public site. On a salon's, customers see the services and real free times, and pick one from their phone.

A chosen time is held for ten minutes so nobody else takes it. Then it becomes a request, and a person at the shop confirms it. Until someone says yes, nothing is booked and the books do not move.

A market lane at dusk: shops lit up under coloured awnings, people walking past.

Sample Salon

Haircut · 45 min

Free times

  • 11:00
  • 11:45
  • 2:30
  • 4:15
Held for ten minutes

Request sent

A person at the shop confirms

Example · synthetic data

The everyday work

Built for the counter, on the phone you already have.

Fast at the rush, clear about who did what, and readable when the network drops.

The counter at rush hour: the shopkeeper hands a bag across, phone in his other hand, customers crowding in.
  • Atta 5 kg₹245
  • Tea 250 g₹130
  • Soap₹38
  • Total₹413

UPICashCredit

Example · synthetic data

Counter

Tap items into a bill and take cash, UPI or credit. Each sale is a row in the day book, under the name of the person who made it.

  • Owner
  • Manager
  • Counter staff

Staff

Everyone signs in as themselves. Counter staff see the counter, credit and stock. Reports, bills and suppliers stay with the owner and a manager.

Read-only · as last opened

  • Sales yesterday₹18,240
  • Items running low6

Example · synthetic data

Offline

Installed on a phone, Munshi keeps your last Home and stock list to read. Nothing can be recorded or approved until you are back online.

How much did we sell today?

Aaj kitna bika?

Hindi

Every screen comes in Hindi and English, and you switch with one tap. Voice works in both.

How the guarantee holds

Nothing changes your books without a named person's yes.

This is enforced by the database, not only the app. Here are the two paths a change can take.

An agent tries to write

AgentBook guardRefused

The only way in

  1. Agent
  2. Draft in the tray
  3. A named personApproved
  4. Book guard
  5. Your books
  • Agents cannot write

    Their database role has no write permission at all. They can read, and they can propose.

  • The guard checks every write

    A write gets in only as a person's own command, or as an approved draft from someone allowed to approve it.

  • Books are never rewritten

    Rows are only added. A correction is a reversal, with a reason and a name on it.

  • Numbers need a source

    A figure on screen is read from a stored row. If it cannot be traced, the screen says Cannot verify.

Open a sample shop and try all of it.

Pick a sample shop and a person in it, or make your own private one. No email, no password.

Demo on synthetic data · suppliers simulated · payments recorded, never processed · PayPal sandbox demo

At night under a street lamp, the shopkeeper pulls down his shop's shutter, smiling, a dog asleep nearby.