How to start with AnomalyGuard

AnomalyGuard watches numbers in your databases over time (sales, occupancy, transactions, and similar metrics). When something unusual happens — a spike, a drop, a gap, or a lasting shift — it flags that as an anomaly.

You do not have to read raw detection output. You save filters (views for different roles) and people open them on Home, or pull the same data through the API / CLI.

This page is the shortest path from “we installed nothing yet” to “we see useful anomalies.”

Who does what

Role	Typical tasks
Admin	Install the app, sign-in (Entra ID), connectors, data views, detectors, users, filters, API keys
Analyst / business user	Open Home, review pinned filters, open anomaly details, add comments

Most of the first-time setup below is admin work. Analysts need a signed-in account and at least one data view assigned to them.

What you need before you begin

Have these ready (or ask IT for them):

Somewhere to run AnomalyGuard — Docker, Kubernetes, or Azure Container Apps (or your team’s usual container host).
Microsoft Entra ID (Azure AD) — so people can sign in. Admins and users get app roles there.
PostgreSQL (or your approved backend) for AnomalyGuard’s own data — anomalies, filters, settings. Plan a backup for this database.
Access to your source warehouse — a database account that can read the business tables you want to monitor (and create objects in the schemas AnomalyGuard uses, when the docs ask for that).

You also need a clear first use case: one metric, one or two category dimensions (for example country + channel), and enough history to learn “normal.”

Step-by-step (first success)

Follow these steps in order. Each step links to the detailed guide.

1. Install AnomalyGuard

Deploy the application and confirm it is healthy.

→ Installation & Setup

When this works, you can open the web UI in a browser.

Register the app in Microsoft Entra ID, assign Admin / User roles, and sign in as an administrator.

→ Setup Service Principal & Users

Without this step, nobody (including you) can use the product safely in a real tenant.

3. Connect to your data

Create a connector: a saved, encrypted connection to the warehouse that holds your metrics.

→ Connectors

Test the connection before you continue. A broken connector blocks every later step.

4. Define what to monitor (data view)

A data view says:

which SQL query loads the data,
how dates and values are named,
which categories (dimensions) split the series,
which detectors look for anomalies,
when to load and process (schedule or API).

Create the definition, deploy it (creates the technical tables), enable it, then run a first load and detection.

→ Data Views · Detectors

Tip: start with one data view and a few detectors. Validate with Data Preview before you trust production schedules.

5. Give people access

Assign data views to users so they only see the datasets they should.

→ Users

If a user has no data views, Home and results stay empty.

6. Create a filter and open Home

An admin builds an anomaly filter (time range, data views, sort, optional aggregation and AI summary) and pins it.

Analysts open Home, pick the pinned filter, and review tiles or the summary.

→ Filters · Home

Optional but useful: add Comments when you know why a spike or drop happened (planned outage, campaign, system upgrade).

Grow from evaluation to production

Stage	Goal	Focus
Evaluation	Prove value quickly	One connector, one data view, one filter, a few days of data
Pilot	Real warehouse, limited audience	More detectors, a handful of users, watch Monitoring
Production	Stable operations	User access, API keys, backups, CI/CD with the CLI, optional AI agents

Automating deploy from Git is covered in CI/CD with AnomalyGuard CLI.

Check that it worked

You are done with “first start” when all of this is true:

You can sign in as admin.
A connector test succeeds.
A data view is deployed and enabled, and a load + process job finished (see Monitoring).
At least one filter is pinned and Home shows anomalies or aggregates.
A non-admin user with assigned data views can open the same Home filter.

If something fails, stay on that step’s guide (connector test errors, deploy validation, empty Home = missing pin or missing data-view access).