You can see your data. Acting on it still takes a queue.

A decade of investment bought visibility: dashboards, warehouses, reports. None of it closes the distance between noticing something and doing something about it. That distance is a queue.

The question nobody asksA question that takes the data team a week is a question most people never ask. By the time the answer arrives, the moment that needed it has passed.

The work that never endsMonitoring, reconciliations, contract checks, audits. Done by hand, one at a time, by people who have other jobs. No team can watch everything, so most things go unwatched until they fail.

The knowledge that walks outThe people who know why the numbers behave the way they do retire, move on, get promoted. Their reasoning leaves with them.

Mentor runs the query and shows the work.

Mentor does not recall an answer or summarise a document. It writes and runs queries against your governed data, over millions of rows, and shows the query, the tools and the source. If you doubt a figure, ask for the underlying rows.

Ask for something the data does not support and Mentor says so rather than producing a number.

MentorProcurement workspace

Top ten vendors by purchase-order spend, last twelve months.

SELECT vendor, SUM(po_value) AS po_spend FROM purchase_orders WHERE po_date >= DATE '2025-09-10' AND po_date < DATE '2026-09-10' GROUP BY vendor ORDER BY po_spend DESC LIMIT 10;

Rank	Vendor	PO spend
1	Northline Industrial Supply	$1,845,619,827
2	Harbor Valve & Fitting	$985,000,000
3	Meridian Chemical Co.	$566,511,713
4	Castell Logistics	$554,743,395
5	Orsa Engineering	$529,966,000

Computed from purchase orders in place. Show the rows

Crews that watch, argue, and speak only on consensus.

A mission is a crew of AI specialists that meets on a schedule, reads its charter, works the data, debates, and votes. They are paid to be critics. Nothing is sent unless the crew agrees a person must act.

Silence is a valid outcome. Most days, that is what you get.

Stability WatchDaily 06:00 · consensus required

SS

Stability Scientist · position

All results posted since last meeting sit inside specification. Study 212-B trends 0.4% below its own history at the 9-month point; still well within limit.

AC

Analytical Chemist · challenge

The 212-B dip coincides with a method change on the HPLC. Method variability, not product. I would not escalate.

QA

QA Skeptic · challenge

Agree, but the next 212-B pull is due in nine days. If it drifts again after the method is stable, that is a different conversation.

M

Mentor · chair

Motion: no action required today. Re-examine 212-B at the next pull. Vote.

No action required. 4 of 4.Nothing sent

Agents propose. People decide.

Nothing consequential happens because an agent said so. It waits, with its reasoning attached, for a person to ratify. Every decision links to the meeting that produced it, post by post. Every action is simulated first, then recorded with its cost, its outcome and an undo.

Needs you1 to ratify

Batch Release WatchScheduled meeting, today

Hold the next cycle on the Line 3 lyophiliser until the port-2 transducer is calibrated.

Port 2 reads above port 1 at steady state, and the offset has grown on each of the last five cycles. Product temperature stayed below the collapse threshold throughout; the batch is releasable. The risk is the next one.

Undo

Talk, code, or your own IDE.

The business user and the data scientist use the same platform and land in the same place. A model built in a notebook is what Mentor answers with; a workflow built from chat is what the engineer sees on the canvas.

Studio: a Python notebook with a running kernel and Mentor beside the code. — **Studio.** A notebook with a kernel, a project per workspace, Mentor beside the code, one-click deploy to a pipeline.

Talk

Ask Mentor to build it: a workflow, an alert rule, a live dashboard, a mission, a model. It authors, validates and deploys, and shows you what it made.

Code

Open Studio: Python notebooks in the project's own environment, import sentinel to reach every source directly, scheduled runs, a model registry and pipelines to deploy into.

Your IDE

Connect Visual Studio Code or the editor of your choice and work on the same projects, the same data and the same deploy path, from where you already work.

Email is a client.

Forward a question, a supplier email or an attachment to Sentinel's mailbox. A thread opens, the analysis runs, and the reply lands back in your inbox. Reply to the reply and the thread continues. Anyone on the CC line joins.

A person who never opens the App still gets full value: ask by email, receive by email, approve by email.

Built for the operation.

The equipment, the work that keeps it running, the supply behind it, and the money it moves. One element tree, the same missions, the same ledger, from the plant floor to the purchase order.

OperationsLive signals from equipment, process excursions, downtime and shift handover. A crew watches every asset of a kind and speaks only when one drifts.

Maintenance and reliabilityWork orders against the signals that justify them. Failure modes per asset. Spares against the plan. Every anomaly reviewed by a crew before a person is asked to look.

Supply chain and procurementPurchase orders, receipts, invoices and agreements across your ERPs. Price paid versus your own median for the same item. Vendor concentration, renewals, and invoice audits from a photo.

Quality and complianceDeviations by age, out-of-trend results, audits and permits, computed from the system of record rather than assembled by hand.

One typed model of what you run.

Raw data arrives as columns and tags with cryptic names. The element layer turns them into the nouns your business uses — a site, a line, a compressor, a purchase order — and hangs every measurement, event and document on the thing it belongs to. Mentor, missions and modules reason about elements, not columns.

Keep your stack.

Sentinel does not replace your systems of record, your warehouse or your BI. It reads them in place and becomes the layer where people and agents act on what they show. Your system of record stays the system of record.

Data in, read in place

ERPs, CMMS and EAM, MES and LIMS. Historians, SCADA and live signals. Warehouses, lakehouses and data platforms. BOMs, drawings, documents and spreadsheets. Email and attachments. Anything with an API.

Actions out, with a record and an undo

Email, chat and meetings. Tickets and work orders. Document systems. Reports as PDF and Excel. Webhooks, storage and transfers. Your own API.

Agents, both ways

Your own agents connect over MCP as the signed-in user, with that person's permissions. Sentinel's agents reach out through every tool above: simulated first, ledgered, gated by the autonomy you set.

Examples, not a list. Connectors and tools are added per deployment.

Arrive Monday. Leave Friday with the problem solved.

Bring one problem that costs you money, the data behind it, and two people who know it. We work it on your data, in your environment, for a week. What you leave with is the problem solved, the platform configured for it, and a production roadmap.

MondayConnectPoint Sentinel at the sources. Read in place. Build the element tree with your team.

TuesdayAskType the questions your team has meant to ask for a year. Get them answered from the data, with the query shown.

WednesdayWatchDescribe the thing you worry about. A mission starts meeting on it.

ThursdayActReports that write themselves. Approvals that wait for a person. Actions with a ledger and an undo.

FridayReviewThe problem, solved, in front of the people who own it. Then a conversation about what should be answerable next.

Start a week Or start with a spreadsheet

On-premises or cloud.

One product, one codebase. It runs inside your own network, in a private cloud, or hosted by Wonder DataLabs. Nothing about the features depends on where it sits, only on where your data is allowed to be.

Behind your firewall

Runs on your own servers or private cloud
Signs in with your identity provider
Reads data that never leaves the network
Models private, or from a provider you choose
Updates delivered as releases

Hosted for you, or in your cloud account

Live in days, not quarters
Same software, same features
Your own tenant, nothing shared
Connects to cloud warehouses in place
Scales with the workload

The audit log, the ledger and every thread live with the deployment.

Safe enough to let it act.

Autonomy is earned in layers, and every layer is visible. Each mission has a ceiling; anything past it waits for a person. Every action an agent takes is simulated first, then recorded with its cost, its outcome and an undo where the tool allows one.

Built by operators who lived the problem.

People who ran the equipment, the shifts and the purchase orders, and got tired of waiting for the answer.

See it on your data.

The practical next step is a conversation about which data you want answerable first.

Start a week Read the brochure