Sense Control architecture.
Describe
Say what you need in plain words — a field, a rule, a report
Preview
Sense shows the exact changes before anything is touched
Approve
Nothing applies until the right person reviews and signs off
Build
Approved changes are made through the platform, as you
Verify
Results are checked, and every step lands in the audit trail
01 · In depth
An admin copilot with your permissions, not ours.
No privileged backdoor
It can only do what you can do
Every action runs as the signed-in administrator through the same APIs, permissions and audit the setup screens use. No super-user, no side door.
Preview-first
See the change before it exists
Records materialize without persisting; rules run in-memory against live data; integration flows dry-run with transforms live and side effects off. Changes are authored inactive, verified, then activated.
Approval gates
Destructive means a dialog, not a hope
Deletes, activations, publishes, imports and bulk writes pause on an in-chat approval — and the copilot asks its questions up front as interactive cards, recommendation marked.
It knows the playbook
Twenty-two task recipes, from entity to sandbox
Curated cookbooks for modelling, layouts, automation, analytics, security, licensing and operations — each with preconditions, steps and verification. When a step needs a human, it says so and gives the exact menu path.
Proof in the chat
The form it built, rendered in the conversation
Live previews of saved layouts and components embed right in the thread; spreadsheets are inspected and preflighted before an import is ever submitted; builds keep streaming while you navigate elsewhere.
And the platform runs itself the same way.
xMatix operations uses the same discipline internally: an ops copilot investigates telemetry around the clock, refutes its own conclusions before raising an insight, and executes only vetted, human-approved remediations.
The capabilities, stated plainly.
Every row ships today and traces to a capability described on this page.
| Capability | xMatix Sense Control |
|---|---|
| Configuration through conversation | Native |
| Runs as the signed-in administrator | Native |
| Preview-first — in-memory dry runs | Native |
| Changes authored inactive, then activated | Native |
| Approval gates on destructive actions | Native |
| Capability | xMatix Sense Control |
|---|---|
| Twenty-two curated task recipes | Native |
| Live previews rendered in the chat | Native |
| Spreadsheet preflight before imports | Native |
| Metered credits and full audit | Native |
The questions evaluators ask.
Assist works on your business records — it answers questions, drafts follow-ups and nudges on the data your teams create every day. Control works on the platform itself: it configures entities, forms, rules, workflows and integrations through conversation. One is a colleague for the people running the business; the other is a copilot for the administrator shaping the system. Explore Sense Assist.
No. Changes are authored inactive, verified, and only then activated — and anything destructive pauses first. Deletes, activations, publishes, imports and bulk writes all stop on an in-chat approval, with the copilot’s questions asked up front as interactive cards and its recommendation marked. Nothing lands in production on the model’s initiative alone.
A working preview, not a description. Records materialize without persisting, rules run in-memory against your live data, and integration flows dry-run with transforms live but side effects off. The form it built renders right in the conversation, and spreadsheets are inspected and preflighted before an import is ever submitted — you approve what you have already seen working.
Changes are staged so that undoing is rarely the question: everything is authored inactive and only activated on your approval, so the deliberate step is the activation itself. Beyond that, platform change management promotes configuration with undo — the same discipline that governs human admins governs the copilot. See how changes promote with undo.
Administrators — and each one only within their own permissions. Every action runs as the signed-in administrator through the same APIs, permissions and audit trail the setup screens use. There is no super-user identity and no side door: if an admin could not make a change by hand, Sense Control cannot make it for them either.