Your technology partner · studied before it is built · run by us

We read your PDFs
and answer your
customers’ questions.

10× the answers. Same team. Catalogues, contracts, spec sheets, precedents. We build the AI employee that reads them all, answers the moment someone asks, and shows the page it came from.

See the worked examples Book a study
LG Your precedent bank Legal · study, then build
RT Your product claims Retail · running today
DS Your catalogue Distribution · study, then build

étoffe noun · French

The cloth a thing is made of. And in avoir l'étoffe de — to have the makings of. One employee is a thread. Enough of them working the same process is a fabric.

Today

The answer exists. Finding it is the job.

A day inside a firm that holds more documents than anyone can hold in their head. None of this is legal work. All of it is billed as if it were.

01

Search the drive for a clause you know you have used before

02

Open six matters to find which version was the final one

03

Ask the associate who drafted it, and wait

04

Rebuild the same summary someone already wrote last year

05

Answer the client question you have answered eleven times

06

Read forty pages to confirm one date

07

Check the precedent bank, find nothing, start from a blank page

08

Explain to the partner where the number came from

After

Every one of those is a question your own documents can already answer. The employee reads them, replies in seconds, and shows the passage it took the answer from — so the person whose name goes on the advice can check it in one click rather than trusting it.

What you get

The answer, and the receipt.

Not a tool your team has to learn. A colleague that already read everything you have. Here is what changes on the first day it runs.

Seconds, not twenty minutes
The question your engineer used to open six files for is answered while the customer is still typing.
The page it came from
Every answer carries its source, so anyone can check it in one click instead of trusting it. Disputes end with a document, not an argument.
The same answer every time
The same question from two customers on two days gets one answer. Nobody is quoting from memory or from an old price list.
"I do not know" when that is true
It answers from your documents or it says it cannot. A confident wrong part number is more expensive than no answer at all.
Open at 2am, in their language
Wherever people already ask you — your site, WhatsApp, Telegram, email, or inside your own tools. Nobody waits until Sunday.
More work, same team
Your people stop retrieving and go back to deciding. You take on more customers without another hire to cover the questions.

Worked examples

One capability. Different documents.

The job is the same every time: read what the business already holds, answer the question, show the line the answer came from. The one marked in brass is running in production today.

RT

Retail · running

The claims your pages make

  • Reads every product page you sell from
  • Flags the claims with nothing behind them
  • Quotes the rule each one runs into
  • Rewrites the weak lines for you to approve

Asked by: ad reviewers, shoppers, your own team

Running now — see a real check

 LG

Legal

The precedents you already wrote

  • Finds the clause you have used before, and the final version of it
  • Flags where a draft departs from your own playbook
  • Answers questions over executed matters
  • Shows the passage behind every answer

Asked by: associates, partners, clients

Studied first · one week

 DS

Distribution

The catalogue in your head

  • Answers what you stock, at what price, in which warehouse
  • Suggests the equivalent part when the first is unavailable
  • Captures the requirement as data, not a chat log
  • Prepares the quotation from your own catalogue

Asked by: customers, on whichever channel they use

Studied first · one week

14

claims found across two live skincare product pages

12

of those the merchant could not evidence

0

claims invented by the agent — every line traces to a source

Measured output from the claims agent, the one role running in production. The other examples carry no numbers yet, and will not until a study and a build produce them.

The obvious question

Why not just use a general LLM?

We do use one — wired to your documents, with a check on every answer. On its own it has no access to your files, no memory of what it told the last customer, and no way to tell you it is wrong.

It invents part numbers
Ask for an equivalent and you get one, confidently, that may not exist or may not match on the spec that matters. Nobody finds out until it is on site.
It cannot show you page 287
An answer with nothing behind it. When a customer disputes a quotation or a partner asks where a figure came from, "the model said so" is not an answer.
It does not know today
Prices moved, stock changed, the spec was revised. A pasted PDF is a snapshot. Your business is not.
Your customer will not use your account
They ask where they already are. A chatbot you log into does not answer the message that arrived on Friday night.

After month one

What stays true after the build.

Models shift, catalogues change, and an edge case nobody scoped turns up on a Tuesday. We take that on, and it is why the numbers hold.

Watched every run
Every action logged with the evidence behind it. You read what it did, not a summary of what it meant to do.
Fixed when it drifts
Providers change models without asking. When output moves, correcting it is our work, not a change request to you.
Updated when you change
New products, new policy, a new approval step. The employee is re-fitted to the process rather than left behind it.
A named person
One engineer who knows your setup, reachable directly. Not a queue and not a shared inbox.
Reported monthly
Against the baseline taken during the study. If a role stops paying for itself, you will hear it from us first.

Industries

Where a workflow is the bottleneck.

The work we take on is repetitive, evidence-bound, and expensive when it is done wrong. That shape shows up in the same six places.

Healthcare

Documentation, coding and discharge paperwork that clinicians do after hours.

Logistics

Vendor quotes, delivery chasing and exception handling across too many inboxes.

Retail & D2C

Product copy at catalogue scale, and the claims inside it that get ads rejected.

Manufacturing

Purchase requests, supplier comparison and the paperwork behind each order.

SaaS

Support volume that scales with customers, and the knowledge base behind it.

Financial services

Reconciliation and month-end close, where the working has to be inspectable.

How it works

Studied before it is built.

01

We study the workflow

One week alongside the people who do it now. We map every step, exception and sign-off, and take the baseline.

02

You get the study

The map, the baseline, and a straight answer on whether a role is worth building here — including no. Yours to keep either way.

03

Then we build

Quoted only after you have read the study. Built against that map and wired to the systems you already run.

04

It runs on live work

One team, one process. Every action logged with the evidence behind it, measured against the baseline from step one.

05

We keep it working

Monitored, corrected and re-fitted as your process changes. The next role starts from one your team has already seen work.

Engagement

Priced per role, not per seat.

You are buying a role that does the work and an engineer who keeps it working. Every stage is paid, including the one where we tell you not to build. We take one client per sector at a time, so the answer is sometimes no.

Study

Fixed feeOne workflow · one week · paid in advance

We watch one process before anyone builds anything, and you keep what we write down — whatever you decide afterwards.

  • Every step, exception and sign-off, mapped
  • The baseline any later claim gets measured against
  • Whether a role is worth building here, including no
  • If yes, what it would cost and how long

Build

Fixed feeQuoted after the study · paid in advance

Built against the map, wired to the systems you already run, then put on live work with your team watching.

  • Nothing integrated until you have read the study
  • Every action logged with its evidence
  • Measured against the baseline we took
  • If the build runs long, that is ours to absorb

Managed

On applicationPer role or per department · monthly in advance

The role stays ours to run. You get the output, the evidence behind it, and someone answerable when it drifts.

  • Monitoring, correction and re-fitting
  • A named engineer who knows your setup
  • Monthly report against the study baseline
  • Additional roles added at a lower rate

Fees are quoted in US dollars, exclusive of any applicable taxes. Every engagement is prepaid — the study and the build in full before work starts, managed roles monthly in advance. There are no invoices in arrears and no lock-in: stop at the end of any paid month, or after the study, owing nothing further. Both the study and the build are fixed-fee, so an overrun is ours to absorb, and we do not bill for roles we have not built yet.

Start with a study of one workflow.

Tell us which process costs you the most when it goes wrong, and we will come back with what a one-week study would cover and what it costs. Want to see the work first? Send a store link and the claims agent — the role already in production — will read it and show you what it found.

Book a study