Ask your systems. Get one answer.
Your client records live in one system, medication in another, training in a third, and none of them talk. Sentinel reads across all of them and answers in plain English. No reports to run, no new logins, and no waiting on the one person who knows where things are kept.
Already a customer? Log in to your dashboard
Ask in plain English
Canada onlyWhose first aid expires before the March inspection?
Watching for you
Three systems, one answer. Read-only, and nothing goes to an outside AI company.
The problem
The answer already exists. It's just spread across five systems.
Nobody can answer it in one place
“Which staff at the Elgin Street home are overdue for medication training?” means opening three systems, exporting two spreadsheets, and cross-referencing by hand. So nobody asks until an inspection forces it.
The person who knew the file is leaving
Ontario’s DSW training pipeline is contracting. When an experienced supervisor leaves, the knowledge of where things are kept and what an inspector actually asks for leaves with them.
You find out at the inspection
Expired certifications, overdue support plan reviews, gaps in a medication record. All knowable in advance, all typically found by someone from the Ministry rather than by you.
Why not ChatGPT
A general AI tool has never seen your records.
It can write you a policy. It cannot tell you whose first aid expires in March, because it has no access to your systems, and giving it that access would mean sending client information to a US company. ECRI named AI chatbot misuse the #1 health technology hazard of 2026. That is exactly the product we did not build.
Connects to the systems you already run
Reads only the tables you approve
Never writes to your records
Runs on Canadian servers only your agency uses
Connecting your data
Your data stays where it is.
No migration, no data dump, no opening a hole in your firewall. A small connector runs on your side and reaches out to us, so Sentinel can read the records you approve without anything being copied out of your network.
Install the connector
A single 5MB program on Windows or Linux, no dependencies to install. It opens an encrypted outbound tunnel to us. Nothing needs to accept incoming connections, which is usually the sentence that ends the conversation with IT.
Choose what it may read
Point it at your database and Sentinel maps the schema automatically. You then set which tables it is allowed to read and which are off-limits. Payroll and HR files can stay invisible to it.
Ask your first question
That is the whole setup. Your team asks in plain English and the answer comes back with the systems it came from, so anyone can check the work.
Data sources
Read-onlyHome Portal
18 tables approved · verified 2m ago
Pharmacy system
6 tables approved · verified 4m ago
Training register
Uploaded · refreshed weekly
Encrypted tunnel, outbound only
No inbound firewall rule needed
Your tunnel is isolated from every other agency
What your team does with it
Ask it, or let it watch for you.
Three ways it gets used, in roughly this order once a team settles in.
Ask
Plain English, answered from every connected system at once. Every answer shows which systems it drew on, so a supervisor can verify it the same way they would verify a colleague.
Whose first aid expires before the March inspection?
Which support plan reviews are overdue, and by how long?
Show me every PRN administration at the Elgin Street home last month.
Which of our homes has the most incident reports this quarter?
Watch
Tell it once what matters. It tells you when a certification is about to lapse, a review is overdue, or a record has a gap, while there is still time to fix it instead of explain it.
Act
Set the routine work up once and it runs on its own. The Monday summary, the reminder to the supervisor, the follow-up nobody has time to chase. Anything that touches a person gets approved by a person first.
Schema map
See what you actually have.
When Sentinel maps a database, it draws it. Every table becomes a building sized by how much is in it, and the links between them show which records are connected to which. Most agencies have never seen their own data laid out this way, and it tends to be the moment the gaps become obvious.
- Every table, its columns, and how many rows it holds
- The links between records, drawn rather than described
- Tables you have marked off-limits shown greyed out
- Useful evidence for a privacy impact assessment
Schema map
Home Portal24 tables · 9 shown · height by row count
Struck-through tables are ones you have placed off-limits. Sentinel cannot read them, whatever it is asked.
Privacy & audit
Built to survive your privacy officer.
Under PIPEDA and Ontario's PHIPA, you have to be able to show where personal health information went and who saw it. That is a reporting problem for most AI tools. Here it is the architecture.
What we can prove, not just promise
Read-only, always. Sentinel has no ability to write to your systems. It cannot alter a medication record or a support plan, so it can never be the reason one changed.
Every query logged. Who asked, what was asked, which tables were touched, and when. Immutable, exportable, and written for someone reviewing it a year later.
You set the boundary. Table-level allow and block lists per data source. If Sentinel was never granted a table, no question can reach it.
Canada only. Inference runs on Canadian infrastructure only your agency uses. No cross-border transfer, and your data trains nobody else’s model.
Audit log
Immutabletraining_records, staff
med_admin
cert lapse · 30d
Home Portal
payroll blocked
Exportable for a privacy impact assessment or a Ministry review.
Who asks it what
Every role gets a different question.
Same assistant, same data, scoped to what each person is allowed to see.
Executive Director
“Are we ready if the Ministry walks in next week?”
One view of where the agency actually stands, without asking four people to build it first.
Director of Quality
“Pull me everything on this individual from the last 12 months.”
Records, medication history and incidents assembled in one pass instead of a week of collating. Scoring your agency against Reg 299/10 is Meridian's job; Sentinel is how you interrogate what it finds.
DSP Supervisor
“What happened at Elgin Street on the overnight shift?”
The shift in one answer, drawn from notes, medication records and incidents together rather than three screens.
IT & privacy lead
“What has it read, and who asked?”
A connector that opens no inbound ports, table-level limits you control, and a log that answers the question directly.
FAQ
Frequently asked questions.
Can it change our records?
No. Sentinel connects read-only and has no write path to your systems. It cannot edit a medication record, close an incident, or update a support plan. When it drafts something, a person reviews and files it in your system themselves.
What can it connect to?
Any PostgreSQL or SQL Server database, including the one behind most agency systems, plus CSV exports for anything that has no database we can reach. If a system offers an API, we can usually read it. Tell us what you run and we will tell you plainly what we can and cannot see.
Do we have to open a firewall port?
No, and this is usually the question IT cares about most. The connector runs on your side and makes an encrypted outbound connection to us. Nothing on your network has to accept incoming traffic. Your tunnel is isolated from every other customer's.
Does it replace Therap, ShareVision or AlayaCare?
No. Sentinel sits on top of whatever you already run and reads across it. Your system of record stays your system of record. Replacing it would mean a migration, and no agency wants a migration.
How is this different from Meridian?
Meridian is the compliance officer: it scans your Home Portal data against Ontario Regulation 299/10 and tells you where you stand before an MCCSS inspector does. Sentinel is the assistant: it reads across every system you connect and answers whatever you ask. Agencies commonly run Meridian for the scoring and Sentinel for everything the scoring raises.
Is this a chatbot for our website?
No. Sentinel is for your staff, not the public. It answers questions about your own operations using your own data, which is the opposite of a public-facing bot answering strangers from a page of FAQs.
Who can see what?
Two layers. You decide which tables Sentinel may read at all, and within that, each person only sees what their role permits. A DSP supervisor does not get the Executive Director's view. Every query is logged either way.
How long does it take to get running?
Connecting a database and mapping the schema is typically a single session with your IT contact. What takes longer is agreeing which tables are in scope, which is a conversation worth having slowly.
What does it cost?
It depends on how many systems you connect and which parts of the platform you need. We will scope it against your agency rather than quote you a number that turns out to be wrong. Ask us and we will be specific.
Get started
Let's talk about your operation.
Tell us what you're trying to solve.