Parts we have built, ready to plug into an ERP, a CRM, an
HR system or a shared inbox. The model reads and writes, code checks
the facts, and a person approves every change.
Prices, quantities, dates and totals are compared by plain code.
Models read the wording and write the words.
A person approves changes
Every change is previewed first and can be undone. Nothing is sent,
deleted or paid for by the AI.
Each block has a test set
Every part ships with fixed questions or examples and a measured
result, rerun before each change.
Your cloud or your office
Hosted, or on a small machine on your premises when messages must
not leave the building.
01Building blocks
The building blocks
Pieces of a custom integration, fitted to the systems you already
have. Pick an area to see where each part fits.
10 building blocks
Message-to-record checker
A second pair of eyes on every reply
Every message is checked against the record it should match. Short deliveries, price rises, late dates and new charges are caught the day they arrive.
Fresh test set: 35 of 37 errors caught, no false alarms
Where it fits
PurchasingSupplier confirmations against the purchase order
FinanceInvoices against the order and goods received; bank-detail changes against the supplier record
SalesCustomer orders against the price list and stock
HRCandidate acceptances against the approved offer; leave requests against the allowance
ServiceCustomer claims against delivery records
Screen-aware assistant
Your staff ask the system in plain words
Questions about customers, orders, invoices or procedures are answered from your own data in seconds, with the list one click away. Nobody waits for a report or hunts through menus.
Question type identified: 87 of 87Right help topic chosen: 31 of 31
Where it fits
Older systemsBuilt-in help for an in-house system nobody documented
Sales"Which customers ordered less this quarter?" with the list one click away
Finance"Which invoices are more than 60 days overdue?"
ServiceAgents look up a customer's last orders without leaving the ticket
HRPolicy and procedure questions answered from the staff handbook
Preview, then apply
AI does the groundwork, people keep the decision
The AI prepares changes to your records and shows exactly what will happen. Nothing changes until a person approves, and every change can be undone.
Up to 100 records per AI-prepared change, with undo
Where it fits
PurchasingAccept a supplier's new delivery dates on the order, with undo
FinanceApply a new price list after reviewing every changed price
SalesMove the ticked deals to the next stage, tasks included
HRChange contract hours for a team once the change is approved
Older systemsSafe bulk updates in an old system without direct database edits
Connector for any AI assistant
Your system, working inside the AI tools your team already uses
Claude, ChatGPT and other assistants can read and work with your system under exactly the same rules as its screens, without replacing it.
Bulk preview of up to 500 records, applied as one undoable change
Where it fits
Older systemsMake an old in-house ERP usable by AI without replacing it
SalesAsk an AI assistant to plan this week's follow-ups from the CRM
FinancePull an aged-debt list into an assistant to draft payment reminders
PurchasingList open orders past their confirmed delivery date
Decision layer
Cheaper, faster AI that knows when it is unsure
A small, fast model makes the routine calls, such as where a message goes or what a reply means, each with a confidence score. Your systems act only when it is sure.
Cost per how-to question down 41%Planted errors caught by the answer check: 16 of 16
Where it fits
PurchasingSort supplier replies into confirmation, change or rejection
ServiceRoute emails and tickets to the right team
SalesRead whether a reply means interest, a no or an opt-out (24 of 24 in tests)
HRClassify job titles by seniority (30 of 30 in tests)
Documents into records
Paperwork into clean records, without retyping
Business cards, signatures, supplier documents and spreadsheets are read into your system, with duplicates found before anything is added.
Иван Петров is found as Ivan Petrov
Where it fits
PurchasingNew suppliers set up from their own documents, duplicates checked
FinanceSupplier price lists in any layout, matched to your item codes
SalesBusiness cards from a trade show straight into the CRM
HRSpreadsheets kept beside the HR system, imported with every change previewed
Live speech assistant
Every call written up, with help while it happens
Both sides of a call are transcribed live, help appears on screen while you speak, and the notes are written in your own template the moment it ends.
Text on screen about 0.7 s after it is spoken60 languages
Where it fits
SalesPhone orders turned into draft sales orders
PurchasingPromises made on supplier calls checked against the order
HRInterview support and candidate assessments in the company's template
ServiceCall notes and follow-ups written in the team's own format
Research at scale
Know who you are dealing with before you commit
Companies and people are researched on the web against your own brief, and the findings go into your records once you have reviewed them.
A drafted first message for every researched lead
Where it fits
FinanceBackground on a new trade customer before setting a credit limit
PurchasingA shortlist of alternative suppliers for a product
SalesResearched prospects with a first draft in your own voice
Health engine without a single score
See which customers and suppliers are slipping, early
Early warning from the data you already hold, with each risk turned into a next action. A strong signal never hides a weak one.
Too little history counts as unknown, never as healthy
Where it fits
SalesTrade customers quietly ordering less, read from order history
PurchasingSupplier health: late deliveries, price rises, quality problems
FinanceCustomers whose payments are starting to slip
ServiceAccounts with rising complaints ahead of renewal
Training videos from the screens
Training that keeps up with every release
Narrated how-to videos are made from your own screens and rebuilt whenever they change, so training always shows the system people actually use.
Videos rebuilt from the script after each release
Where it fits
Older systemsHow-to videos for an in-house system, rebuilt with every release
HROnboarding videos for new starters
SalesProduct walkthroughs for customers
02Combinations
Blocks combine into solutions
The same parts, put together for one problem.
Supplier replies, handled
Every supplier reply checked against its order. Buyers see what
changed, accept it with one click and can undo it. Phone promises go
through the same checks.
Message-to-record checker
Decision layer
Preview, then apply
Live speech assistant
An AI-ready system without replacing it
Built-in help and "ask your system" for an older in-house ERP, a safe
connection for AI assistants, and training videos that stay
current.
Screen-aware assistant
Connector for any AI assistant
Training videos
Customers and suppliers at risk
Early warning from the data you already hold: customers ordering
less, suppliers slipping, payments drifting. Each with a next step
and background research.
Health engine
Research at scale
Screen-aware assistant
03How it runs
How a project runs
Small steps, each with a decision point.
Step 1
Proof on your data
One to two weeks on a sample of your real messages and records. You
see the catch rate and the false alarms before you commit.
Step 2
Pilot in the background
Alerts go to one person. Nobody's work changes yet, and results are
reported every week.
Step 3
Go live
On your systems, hosted or on your premises. Changes stay
preview-first.
Step 4
Support
Monitoring, a monthly accuracy report, updates and new formats as
your suppliers and customers change.