Release 2.1 is the biggest step forward since we shipped 1.1. It is not a patch release with a few bug fixes – it is a broader platform story: more ways to catch real problems in your metrics, less noise from expected calendar effects, a sharper analyst UI, hardened security, and a deployment path that fits how Azure teams already work.
If you run time-series monitoring in SQL warehouses and need ranked priorities instead of another dashboard to babysit, this release is worth a closer look.
Deploy on Azure, sign in with Entra
AnomalyGuard 2.1 is available on Azure Marketplace with ARM templates and managed-application packages for Start, Small Team, Growing Team, and Enterprise Team SKUs. You deploy into your own subscription; your data stays in your infrastructure.
Sign-in is integrated with Microsoft Entra ID. During Marketplace deployment you configure tenant and app registration. Inside the product, the Users admin page maps Entra identities to data views, so Home shows only the anomalies each person is allowed to see.
Architecture that scales with your jobs
Behind the UI, the Master + Worker model is unchanged in concept but stronger in practice:
- Master enqueues load and detection jobs.
- N Workers process the shared task queue.
Release 2.1 brings an optimized task queue schema, Docker images for Master and Worker, API hardening (gate middleware, security headers, rate limiting), and cleaner anomaly indexing with an IsCurrent flag so recalculation respects current vs. historical rows.
The goal is predictable operations at scale – not a single box that falls over when you add data views.
A redesigned analyst UI
Day-to-day work should feel faster, not heavier. 2.1 refreshes the screens analysts and admins touch most:
Home
Redesigned result cards with configurable max tiles (5, 15, and more) and Summary / All views – useful for executive scanning or kiosk displays.
Monitoring
List and detail pages are split. The detail view puts load, range, anomaly, and task stats in one row, so you can see job health without clicking through four screens.
Data views
Enhanced YAML editor, load-mode persistence, and a license usage counter ({count} from {limit}) so admins know where they stand against the plan.
Data preview
Dry-run detectors on real loaded data before you trust a cron schedule. You get raw and processed charts, chart-only runs (no detectors), skipped-detector reasons when a series does not qualify, and Plotly chart builder improvements. Tune parameters, reduce false positives, then copy the YAML back into the data view.
Everything as YAML
Configuration in AnomalyGuard is not scattered across opaque UI state. A data view is a versionable YAML document: source query, categories, detectors, special dates, and schedule – one file you can review in Git, apply from a pipeline, or edit in the UI.
Typical workflow:
- Author or generate YAML locally or in the editor.
- Commit to source control.
- Apply via UI, REST API, or
anomalyguardCLI. - Deploy, enable, and enqueue load / process jobs from CI/CD.
Special dates live in the same model – explicit dates, recurring rules (month_end, month_start), and year-crossing periods – so known calendar effects do not flood your anomaly cloud.
Twelve new detector types
Release 2.1 adds a full catalog of detectors aimed at real business monitoring -not just “something moved yesterday.”
All of them are configured in data view analysis YAML, with shared behavior improvements: direction mode (above | below | both), aggregate mode (sum | average on weekly/monthly detectors), series filters with where and maxValue, and baseline precomputation for peer, mix, and cardinality detectors at load time.
Data quality and freshness
| Detector | What it catches |
|---|---|
| Bounds | Values outside [min, max] |
| Stale | Last point older than your freshness SLA |
| Flatline | Near-zero or stuck-constant runs |
| Zero rate | Too many near-zero days in a window |
Trend and seasonality
| Detector | What it catches |
|---|---|
| DoPY | Today vs. same calendar day last year |
| Sustain | Multi-day run above or below baseline |
| Seasonal residual | Outliers after trend + seasonality |
| Volatility | Regime shifts in rolling variability |
Business and structural comparisons
| Detector | What it catches |
|---|---|
| Target | Actuals vs. explicit targets (with per-dimension scope) |
| Peer | A series vs. peers in the same group |
| Mix | Unusual share of a child in a parent aggregate |
| Cardinality | Unusual distinct-value counts in a group |
Each detector has its own page in the documentation with parameters, examples, and diagrams.
Special dates: less noise, more signal
Retail calendars, month-end closes, and seasonal campaigns create patterns that look like anomalies if you treat every Tuesday the same.
With useSpecialDates: true on a detector and rules in YAML (specialDates, recurringRules, specialPeriods), matching anomalies are suppressed based on direction and configured impact (positive, negative, both).
You still see real breaks. You stop drowning in expected ones.
Connectors: Fabric and structured config
Microsoft Fabric SQL joins SQL Server, PostgreSQL, Oracle, and Snowflake. The Fabric connector uses service-principal authentication via structured config (server, database, tenant_id, client_id, client_secret).
Connector settings are stored as encrypted JSON – the UI adapts fields per engine. Legacy API clients can still send connectionString; it is converted on save. Query preview for connector sampling is improved for data preview runs.
Filters, comments, and LLM summarization
Saved filters are a first-class admin surface with Aggregates, Anomalies, and Summary tabs. Pin filters per user, share them across the team, and see comment indicators on dates with operational context.
LLM summarization is configurable in Setup (global templates) and per filter. Refresh summaries on demand or when underlying data views change. Settings are redacted in API responses.
The same filter an analyst pins on Home can become an executive brief – or the input to an automation.
CLI, API keys, and CI/CD
Operations teams should not click through the UI to promote config.
anomalyguardCLI – cross-platform, downloadable from Master, aligned with the app version.- API keys – scoped to data views, with optional descriptions for pipelines and agents.
Example pipeline steps:
export ANOMALYGUARD_API_KEY=$PIPELINE_KEY
anomalyguard dataviews create --file dataview.json
anomalyguard dataviews deploy 5
anomalyguard dataviews enable 5
anomalyguard dataviews jobs-load 5
anomalyguard dataviews jobs-process 5Docker images and Azure Pipelines builds (ACR + docs publish) round out the packaging story for teams that deploy AnomalyGuard as infrastructure, not a side project.
AI across the product — not a bolt-on
AI in 2.1 is woven into existing workflows:
- Data view YAML – generate and refine prompts to draft or change detector config.
- Filter summaries – LLM narratives for leadership, on demand.
- Agents – the same saved filters and
--jsonCLI output that humans use, without opening the UI.
People, pipelines, and agents share one source of truth: YAML config, deployed data views, and named filters.
Security and operations
Release 2.1 hardens the platform for production API exposure:
- Centralized credential encryption and SQL query safety checks.
- Per-data-view secret keys removed – access flows through standard auth.
- Consistent UiLog error reporting across Blazor pages.
Combined with Entra sign-in and scoped API keys, the security model is simpler and easier to audit.
Documentation that matches the product
We rebuilt the docs site at docs.anomalyguard.com with Admin Guide, User Guide, Tutorials, and Release Notes.
New in 2.1:
- A Detectors catalog with a dedicated page per detector type.
- Updated Connectors, Data Views, and Data Preview guides.
- Sample YAML covering all new detector types.
If the product moved, the documentation moved with it – including the examples you copy into production.
Get started
- New customers: deploy from Azure Marketplace and follow the Admin Guide for Entra setup.
- Existing customers: upgrade to 2.1, review the Release Notes, and explore the new detectors in Data preview before enabling schedules.
- Platform teams: pull the CLI, issue an API key, and wire data view deploy into your pipeline.
AnomalyGuard 2.1 – pure business signal, defined as YAML, deployed on Azure, explained in documentation you can actually use.
Questions or feedback? Reach us through anomalyguard.com.