Your people already have an assistant open. A sales manager drafts the quarterly review in Claude. An analyst works through a question in ChatGPT. An administrator lives in Claude Code. Until now, the business sat in another tab: copy the numbers out, paste them in, hope nothing changed in between. Sense closes that gap. Connect the assistant you already use to your xMatix organisation once, and it can ask your records, run your reports and dashboards, search your knowledge and, if you allow it, act — as you, inside your permissions, with every call audited.
What the assistant can do once it is connected
The connector exposes the same tool surface the in-product Sense agent uses, so nothing is special-cased for the external assistant and nothing is weaker. It discovers your entities and their fields by name, queries and aggregates records, walks relationships, runs the saved reports and dashboards your business already trusts, and searches the knowledge collections that ground your Sense agents. Writes are a separate scope your organisation grants deliberately.
It runs as you — here too
The rule that governs every Sense surface does not bend for an external client. Every call is evaluated three ways, and each layer can only narrow what the one before allowed:
- The token's scopes. Read, write and setup are separate. A write scope is only ever issued when the organisation has allowed writes; if it has not, the assistant never sees a write tool, whatever it asked for.
- The organisation's capability grants. A tool a Sense agent could not be granted in AI Studio is not offered to an assistant either. The external surface is a subset of the internal one, never a superset.
- Your own permissions, at execution time. Security profiles, record security and restriction rules apply to every query exactly as they do on screen. There is no service account behind the connector answering from data you could not open yourself.
Tokens are bound to the organisations you approved and expire after an hour; refresh tokens rotate, and a replayed one revokes its whole family. Every tool call is audited as you. How Sense earns trust describes the contract in full; the connector inherits all of it.
Off until you turn it on
External assistants are disabled for every organisation by default. Enabling them is an administrator's decision, made in Setup → Sense AI Studio → External Assistants, and it comes with three controls: the switch itself, a separate switch that allows writes, and an optional allow-list of approved assistant clients. The same page lists every live connection — which assistant, which user, what access, when last used — with a per-connection disconnect. Turning the switch off ends every session immediately.
| Control | Default | What it decides |
|---|---|---|
| Allow external assistants | Off | Whether any assistant may act as this organisation's users at all. Off cuts existing sessions. |
| Allow writes | Off | Whether the write and setup scopes can be issued. Off caps every token at read-only. |
| Approved assistants | Any | An optional allow-list of assistant clients; everything else is refused at sign-in. |
| Connections | — | Every live connection by assistant and user, with last-used time and a disconnect action. |
Why an open standard, and not a plug-in
We could have shipped an extension for one assistant. We chose the protocol those assistants already speak, for the same reason the platform exposes REST and OData rather than a proprietary client library: your choice of assistant should not be our decision. The same endpoint serves Claude on the web and desktop, ChatGPT's connectors, Claude Code in a terminal, and any agent framework that speaks MCP — and the authorization server in front of it is a general one, so the next integration your IT team wants to build signs in the same way. The developer documentation has the protocol details and the full tool list; the setup guide walks an administrator through the first connection.
Speaking the standard does not stop us meeting people where they are. The same endpoint is packaged as an installable Claude Code plugin, so a terminal user installs xMatix by name instead of typing a URL, and every tool carries the read-only or destructive annotation the assistants use to decide when to ask before acting.
Common questions
Which assistants can connect to xMatix?
Any client that speaks the Model Context Protocol over Streamable HTTP with OAuth sign-in: Claude on the web and desktop through a custom connector, ChatGPT through its connector settings, Claude Code from the terminal (by name through the xMatix plugin, or as a plain MCP server), and agent frameworks built on the same standard. There is nothing to install on the xMatix side; the endpoint is part of the platform.
Can an assistant see data the user cannot see?
No. Every tool call runs as the signed-in user, through the same record security, field permissions and restriction rules that govern every screen. The token also carries only the scopes the organisation allows and only the tools a Sense agent could be granted, so the external surface is always a subset of what the user could already do inside xMatix.
Can it change data?
Only if the organisation has switched writes on, and then only through the create and update tools under the user's own permissions. Administrators additionally receive the setup tools, annotated so the assistant confirms anything destructive with them first. With writes off — the default — every token is capped at read-only regardless of what the assistant requests.
Is our data used to train the assistant's model?
xMatix does not train models on your data, and the connector does not change that. What an assistant provider does with the content of a conversation is governed by your agreement with that provider, exactly as it is when a colleague pastes a spreadsheet into it today — the difference is that with the connector the data arrives permission-checked, cited and audited rather than copied out by hand.
How do we switch it off?
An administrator turns off "Allow external assistants" in Sense AI Studio and every existing session ends immediately. Individual connections can also be disconnected one at a time from the same page, and a user can remove the connector from their assistant at any time.
Does it need a separate licence?
No. A person connects with their normal xMatix login, and because the assistant reasons with its own model the connector consumes none of the organisation's AI credits. Which tools appear follows the organisation's Sense capability grants and licensing for reports, dashboards and knowledge — the same rules as in the product, nothing extra.