Ask your business a question.
UNI4C builds AI that reads the systems you already run, answers with the records behind it, and then does something with the answer: a report, a workflow, a draft, a pull request. Business Central is where it goes deepest.
Question
Which customers are over their credit limit right now?
bc__read_data · Cust. Ledger Entry · 1,284 records
| Customer | Balance (DKK) | Credit limit | Over by |
|---|---|---|---|
| Nordisk Trædesign A/S | 412,900 | 350,000 | 62,900 |
| Vestjysk Maskinfabrik | 288,450 | 250,000 | 38,450 |
| Kolding Byg ApS | 197,200 | 175,000 | 22,200 |
3 of 1,284 entries matched. Read with your own Business Central permissions.
One question, and everything downstream of it
The query that answers a question in chat is the same query that becomes the report you open every Monday, the workflow that mails it on the first, and the definition the assistant remembers is yours. Nothing is retyped between them.
Ask
In your own words, in Danish or English. An unsettled reading is asked about, not guessed.
Which customers are over their credit limit right now?
ControllerRead
A real query against your tenant, through your own connection, paging automatically to 2,000 records.
bc__read_data · Cust. Ledger Entry · 1,284 records
SecurityVerify
Every figure in the reply is checked against the rows that came back. An ungrounded number never reaches you.
3 customers · 124,550 DKK over
How that worksPresent
The rows render as a block (table, value, chart), and the same block can become a report page.
present · table · 4 columns
ReportsKeep
What was settled becomes a reviewed entry, with the query that established it stored as a replayable recipe.
Over credit limit = Balance > Credit Limit, open entries only
Knowledge
It plugs into what you already run
One assistant, several systems: from the ERP to the mailbox to the telemetry behind it. Each connector reads with the permissions of the person asking, never more.
- Cust. Ledger Entry
- Sales Header
- Item
- Job
- Time Sheet
The deepest integration: ledger entries, documents, items, projects and posted time, plus a signed extension that pushes events out of your tenant as they happen.
- Messages
- Drafts
- Mailbox settings
Your own mailbox, read and send, through Microsoft Graph and your own delegated permission. Drafts are the default; sending asks first.
- DebtorClient
- CompanyClient
A second ERP, same rules: the connector discovers the entities your Uniconta plan exposes, reads through your own connection, and every write asks first.
- Workspace
- Semantic model
- DAX
Your semantic models, queried where they live: the assistant lists workspaces, describes a model's tables and measures, runs DAX against it, and can start a refresh.
- AppTraces
- AppExceptions
- AppPageViews
The telemetry your Business Central tenant already emits (errors, slow operations, usage), read from your Application Insights workspace with KQL you can inspect.
- Repository
- Pull request
- AL symbols
Where App Builder works: your own GitHub or a Gitea we host, read repo by repo, every change arriving as a review your consultant signs off.
Five surfaces, one set of connections
Each one reads with the permissions of the person asking, and each one shows what it read.
Controller
Ask in plain language and get an answer grounded in the records your Business Central tenant holds right now, not in a training set, and not in last night's export. In analysis mode every figure in the reply is checked against the rows that were actually read.
- SupportAnswers on the Business Central page the user is stuck on.
- ReportsLive reports, documents and exports built on queries the assistant already ran.
- Dynamic workflowsAutomations that watch Business Central and act when something changes.
- App BuilderDescribe the app you need in plain language, try it in your own sandbox, and go round again until it is right.
Before any of that matters, three answers
The assistant reads real ledgers and real mailboxes. The first questions a finance team asks are not about features, and each one has its own page.
- Will it understand our business?
It learns yours as it works, on its own
Your definition of overdue, of an active project, of revenue. The assistant writes down what gets settled, a second agent checks it, and the next answer uses it. Nobody has to build the layer, and you can still read, edit or overrule every entry in it.
Knowledge- Can it make something up?
It writes the query, not the number
Every figure you see is a block the platform rendered from the rows a query returned, with the tool, the table, the filters and the record count attached. The assistant is told a table appeared, never the values in it.
Security- Where does our data go?
In Europe, and gone when you say
Application, database and models all run in the EU. Each company sets how long a conversation is kept (one day to a year), and the window is enforced, not promised.
Data
See it against your own ledger
A demo on a tenant that looks like yours answers more than any page can. Half an hour, no slide deck.
Or write to us at hello@uni4c.ai