Internal assistant
Answers questions about how your company works from your policies, procedures and past work. Every answer links to the document it came from.
internal AI tools · for teams of 20–200
An assistant that knows how your company works.
Built on your own documents and processes, inside your accounts, and it shows where each answer came from.
You'll talk to me, Lucian
sound familiar?
The answers exist. Finding them takes a person.
- “Everyone asks the same three people.”
- “The answer is somewhere in SharePoint.”
- “We tried ChatGPT. The team loved it. Legal said no company data goes in.”
- “Every new hire asks the questions the last one asked.”
what it can be
Four tools a team can start with.
Each one works from your documents, runs in your accounts and shows its sources. The pilot tells you which one pays first.
Document reader
Reads the documents that arrive every day, such as statements, invoices, contracts or forms, and turns them into data your systems use. DocUp runs one, below.
Quoting tool
Drafts a quote from a request, your price list and your past quotes. A person checks it before it goes out.
Support helper
Drafts replies to customer questions from your help articles and past tickets, with the sources attached. Your team edits and sends.
how it works
Two weeks to prove it, then build, then run.
Free 15-minute call
The questions your team repeats and where the answers live. I tell you whether a tool can answer them. No access needed.
Two weeks: the pilot
$3000, fixed. One team, one document source. A working version in your accounts, tested on 50 real questions your team writes. You see every answer and where it came from.
Build the tool
$9000–20000, fixed in writing after the pilot: more sources, access by role and the tool where your team already works. The pilot’s $3000 comes off the price.
Run it
$750 a month keeps it running as your documents change: sources in sync, answers checked, costs watched. Or you take the documentation and run it yourself.
the price
A fixed price for each step, agreed before it starts.
- Pilot
- $3000
- Two weeks, one team, one document source, tested on 50 real questions. Credited against the build.
- Build
- $9000–20000
- Fixed in writing after the pilot. Where it falls depends on how many sources and teams it covers and what it connects to. Half to start, half when it runs.
- Run
- $750/mo
- Keeps it running and accurate as your documents change. Billed monthly.
US dollars, excluding VAT where it applies. The AI provider and the hosting bill you directly.
for a single task
One repeated task rather than a tool for a team? Each of the ten automations is $800–1500, fixed.
See the menuif the pilot misses
Your questions, your pass mark. Miss it and you stop there.
- The pilot is tested on 50 real questions your team writes, not on questions I pick.
- What counts as a right answer is agreed in writing before it starts.
- If it doesn’t get them right, you stop there and owe nothing past the pilot.
- If you go on, the pilot’s $3000 comes off the build.
data and security
Your documents stay in your accounts.
The reason legal said no to ChatGPT is the first thing the design answers.
- It runs in your cloud account and your AI provider account, under your company’s agreement with them. Nothing runs on mine.
- You choose the AI provider, and your legal team signs off on it before the pilot starts.
- People get answers only from documents they can already open.
- Every answer shows its source, so anyone can check it.
- Every question and answer is logged in your account.
- When we stop, every account, key and document stays yours, and my access is removed.
named client
An internal AI tool already doing the work.
DocUp reads its bank statements with stmtscan, the tool we built for it, for under a cent each instead of 70 cents.
DocUp · internal tool, merchant cash advance
About $60k a month off its costs, and a leaner team
DocUp is an internal tool for the merchant cash advance industry: merchant leads, their bank statements and the lenders they already owe. It was reading about 4,500 statements a month and paying another provider $0.70 for each 10-page one. For DocUp we built stmtscan, which does it for under $0.01. It splits each PDF into pages, reads them with AI and returns the balances, deposits, NSF fees, negative days and every transaction, with payments to other lenders flagged. A statement missing its balances is flagged, never marked done. Together with the other automations, it took so much manual work off the team that DocUp now runs with fewer people, and its costs are about $60k a month lower.
questions
Before you book.
- Can’t we just use ChatGPT?
- For public information, often yes. For contracts, client files and internal procedures, it means pasting company data into a tool your legal team hasn’t approved. This runs in your accounts, on your documents, and shows where each answer came from.
- What if it gets an answer wrong?
- Every answer shows its source, so a wrong one is easy to spot and trace. It is built to say when your documents don’t cover a question rather than guess, and the pilot measures how well it does that on 50 of your real questions before you pay for a build.
- Which documents can it read?
- Wherever your team keeps them, if they can be exported or reached through an API: SharePoint, Google Drive, Notion, Confluence, PDFs, email or your own database. The pilot starts with one source.
- Who owns it?
- You. It runs in your accounts, the code and the documentation are yours, and it keeps working if we never speak again.
- What does it cost to run?
- The AI provider and the hosting bill you directly, in your own accounts. They are not part of the fee.