The AI that works with your data
Five ways of turning a question into an answer that stays useful — a scoped Agent to ask, a dashboard or report that stays live, a Tracker that keeps watching, an Automation that keeps working. All governed the same way: cleared data only, reviewed before it's live, on the record afterwards.
A named AI, built for one job
Give a job its own Agent — a Finance Analyst, an Onboarding Assistant, a Support Triage — with its own instructions, its own model per capability, and skills switched on individually: querying data, reading a knowledge base, building dashboards and reports, running Automations, acting through an approved tool. Nothing is reachable by default, and an Agent is reviewed and granted to a group up to a classification ceiling like anything else that touches your data.
- Its own instructions and model, chosen for the job it does.
- Knowledge bases as an explicit skill — long documents, scanned files, and the diagrams inside them, with sources cited.
- Approved before anyone can use it, the same way a database connection is.
Describe a dashboard, get a live one
Describe what you want to see in plain language and get a working dashboard back — not a mockup. It stays live: refreshed for whoever opens it, showing each viewer only the rows and columns they're cleared for, rather than freezing whatever the author could see when it was built. Charts can be adjusted by hand afterwards.
- Live, not a snapshot. Refreshes for whoever opens it, not just whoever built it.
- Masked per viewer — the same dashboard, correctly redacted for each person's clearance.
- Editable afterwards, like anything else — generated first, then adjusted by hand.
Reports that keep themselves current
Work up a report or a memo in conversation, drawing on your databases and document collections, then publish it. Set it to refresh on a schedule and it keeps the same address as it updates, with earlier versions kept behind it — so a link shared last quarter still works, and still shows what it said then.
- One link, always current — refreshes on the schedule you set.
- Every earlier version kept — nothing a link pointed to quietly disappears.
- Data and documents together — not two separate tools to reconcile.
Standing monitoring, not one-off reports
Describe what you want watched — overdue invoices above a threshold, weekly signups by region, stock cover on your top lines — and it becomes a named check that runs on its own, building a history of how a number moved rather than answering once and forgetting.
- One number, watched closely, building its own history run over run.
- Thresholds you set, so it knows whether a rise is good news or bad.
- Proves itself before anyone relies on it — a clean run end to end before it's shared.
Eight weeks, recorded once each Monday.
A standing AI worker for the jobs nobody has time for
Describe a recurring job in plain language — a weekly summary, a monitoring sweep, a routine check against a website or system — and it restates back what it understood before it's ever left to run unattended. From then on it works to a schedule, drawing only on the web, knowledge bases, and databases you've allowed, and hands back a write-up, a number, or a file depending on what the job calls for.
- Confirmed before it runs unsupervised. A written restatement you sign off on.
- Scoped sources only — web search, knowledge bases, and databases opted into individually.
- Every run on the record — what it read, what it wrote, and what it did through a tool.
All five, governed the same way
Different jobs, one set of rules underneath.
Cleared data only, every time
Whatever an Agent, dashboard, report, Tracker, or Automation touches, it touches within the same clearance and masking rules that apply everywhere else — there's no separate, looser path through an AI feature.
Reviewed before it goes live
An Agent, a scheduled Automation, and a published dashboard all go through the same approval workflow as a new database connection — so "what's this allowed to do" is always answerable.
See these against your own data
Much easier to judge against numbers and jobs you recognise than to describe in the abstract.