An agent that can only act on the record.
Sentinel is a governed AI agent over the same 47 role-scoped tools the console runs. The tools are live today; Sentinel is being brought up on them, assisting first, then acting within guardrails.
In build. The governed tool surface it runs on is live today.
The same tools as your team. Not one more.
An agent on the governed surface
Sentinel runs on the same 47 role-scoped tools the console runs. The tools are live today; the agent is being brought up on them.
A role, not root
Sentinel holds a role like any operator. What it can touch is scoped the same way, which means it cannot do anything your team could not.
Zero actions off the trail
Every agent action is attributed to Sentinel and written to the same immutable audit trail as operator and resident actions. Nothing it does is invisible.
Staged autonomy
Assist first: drafts, summaries and answers from the record. Acting within guardrails comes second, and only inside the same role scope.
Chat first, voice next
Ask questions in plain language over the governed record. Chat lands first, with voice to follow.
One platform. Not a product bundle.
Sentinel is not a bolt-on chatbot with database access. It is a consumer of the same governed tool surface as the console and the API, which is exactly what makes it tolerable in a live building.
The surface it consumes
The MCP server exposes the same governed tools to agents, Sentinel included. It is not a special case.
ExploreWhat makes an agent safe
Role scoping, attribution and the immutable trail apply to AI actions exactly as they do to human ones.
ExploreAnswers from the record
The numbers Sentinel reasons over are the same governed figures analytics reads.
ExploreTrust is structural, not promised.
The honest answer to whether you can trust an AI agent in a live building is that you should not have to. Sentinel cannot exceed its role, every action it takes is attributed and recorded on the immutable trail, and its autonomy is staged: assisting before acting, always within guardrails. The 47 governed tools it runs on are live in the platform today.
Read: governed AI agents in building operationsCan we trust an AI agent in a live building?
Sentinel cannot do anything your team cannot, because it works the same role-scoped tools under the same governance. Every action lands on the same trail, so the question is never what the agent did, only whether you want it doing more.
See the surface Sentinel will run on.
The 47 governed tools are live today and the agent is coming up on them. Book a pilot and see the tool surface working, with the trail behind it.
Prefer email? hello@ark360.com.au