ATX AnalyticsTable Release Notes | What's New in Every Version

What's new in ATX AnalyticsTableEvery ATX AnalyticsTable release, explained for the people who use it: what you can do now, what got better, security and privacy changes, notes for IT administrators, fixes, and anything to know before you upgrade.

Version 1.6.0 — 3 November 2026 — Free to use, one governed number everywhere, and an assistant that tells the story behind it

ATX AnalyticsTable now has a Free edition: without a licence the app stays fully usable, for private and commercial work, within sensible limits. A governed metric is now the same number in every chart, cell, pipeline, alert, API call and embedded dashboard — and you can bring your existing dbt or OSI model. Ask Greg explains what drives a number and what is unusual, writes data stories in which every figure is checked, and builds charts, pipelines and stories on the metrics your organisation agreed on.
A Free edition — no licence needed: Without a licence, or after a trial ends, the app keeps working and stays writable — free for private and commercial use, within limits.
Data stories with every figure checked: A short narrative for a dashboard, written from facts the analysis tools compute. Every sentence cites its sources, and a sentence with a figure the facts do not hold is removed.
One governed number, everywhere: Certified metrics drive charts, pivots, KPIs, a METRIC() cell formula, pipelines, alerts, the API and embedded dashboards — and reports can be held to the certified version.
Bring your dbt or OSI model: Import the semantic models and metrics you already defined in dbt, or an Apache Ossie (OSI) model, with a preview of what is created and what is left out.
Open to your tools — API, MCP and embedding: A documented, read-only REST API; MCP servers in both directions — BigQuery or Jira for Ask Greg, and your governed metrics for outside AI agents; and dashboards embedded in your own portal.
What drives it, and which groups stand out: Key-driver and cluster charts explain a number by its drivers and split your rows into segments — on dashboards, on sheets and in Ask Greg.

Version 1.6.0: good to know before you upgrade

1.6.0 installs over 1.5.0 and keeps your projects, settings and licence. Licensed installations see no new limits.
Installations without a valid licence now run on the Free edition instead of read-only. If such an installation has more connections, schedules, pipelines or dashboards than Free allows, the extra ones are paused (not deleted) until a licence is added — check Settings → License after upgrading.
A dashboard you return to now shows its last numbers for a moment while it refreshes in the background; the spinning mark in an object's title strip tells you it is still checking.
On MySQL, a chart over a text column that holds values such as "n/a" used to count them as 0; they are now left out, so such a histogram or total can change.
Monitoring that reads memory or queue figures from the readiness check must now send the API token, or use the metrics endpoint instead.

Version 1.6.0: New

The Free edition: An installation with no licence, an evaluation that has ended, or a licence past its grace period now runs on the Free edition instead of becoming read-only. It is writable and free for private and commercial use, with up to 2 database connections (the built-in database never counts), 2 scheduled jobs, 2 pipelines, 2 automated PDF reports, 1 dashboard and 3 sheets of 1,000 rows per project. Local databases (DuckDB, SQLite, PostgreSQL, SQL Server, MySQL) and files are included; data warehouses, Oracle and cloud storage need a licence. Anything over a limit is paused, never deleted — Settings → License shows what runs and lets you choose.
Data stories: Dashboard → Data story writes a few sentences about a dashboard or a set of metrics: what changed since the last period and a year ago, which groups drove it, what was unusual, what the drivers are and where it is heading. The analysis tools compute the facts and the AI only words them: every sentence cites its sources, and every figure is checked against the facts it cites. Copy it as Markdown, save it as a document that opens in Word, or add it to a PDF report, where it is written afresh on every run. Without an AI model, or when your data policy sends no figures, the story is written from templates on your computer.
AI explanation boxes in PDF reports: Add an AI explanation box next to any table, chart, pivot or dashboard in a PDF report. It explains that element from its own numbers, in a framed box marked as AI-generated, and flags any figure it cannot find in the data. If the AI is unavailable the report still prints.
Alerts that explain themselves — and "Check with AI": An alert rule can add a short AI paragraph to its e-mail. The rule's condition, never the AI, decides whether it fires; a paragraph with a figure the data does not hold is not sent, and an AI outage never stops an alert. Check with AI tries a rule on today's data without sending anything and tells you whether it would fire, is held back, is close (within 10 %) or is clear — and why.
Insight watches: Be e-mailed when a number does something unusual, moves more than a set percentage against the previous period or a year ago, or is forecast to cross a line — on a table or on a governed metric. The same statistics Ask Greg uses decide, so a watch costs nothing to run.
Key-driver analysis — "what drives X?": The new Key drivers chart explains a number such as revenue, or a yes/no outcome such as churn, by up to 12 columns. Each driver gets a fair share of the explanation that adds up to 100 %, with its direction and whether it is significant, plus the categories and value ranges that stand out — click one to filter the dashboard. It reports association, not cause, and says so.
Cluster analysis — find the segments in your data: The Clusters chart groups rows or customers into segments by up to 12 numeric features, picks the number of segments for you, and shows each segment's size, averages and what sets it apart. The same data gives the same segments on every computer and database. Click a segment to filter the dashboard to its members, or export the members as CSV.
Ask Greg explains change and looks ahead: Ask "is anything unusual?", "what will next quarter look like?" or "how did this month compare?". Ask Greg finds the periods that break their own trend and season, forecasts with the same model as the forecast chart — so the number it quotes matches the chart it proposes — and compares periods with each group's share of the change. It can also run key-driver and cluster analyses on your tables.
A governed semantic layer: Define measures, dimensions and hierarchies once — shared across the installation or kept inside a project — with business definitions, synonyms and owners. Content stewards certify definitions, every change is versioned and can be restored in one click, and a certification lapses when the definition changes. Ask Greg answers metric questions from these definitions first and marks each answer Certified, Governed or Unverified.
Governed metrics in charts, pivots, KPIs and cells: Choose a governed metric instead of a table as the source of a chart, pivot, KPI or table, on a sheet or a dashboard — it follows the definition and warns before an average or a ratio is re-totalled. =METRIC("revenue", "region", "EMEA") puts one governed number in a cell.
Governed metrics in pipelines: The new Governed metrics step (Transform → Pipelines → Add step → Data in / out) reads certified figures into a scheduled pipeline instead of hand-written SQL. A result cut at the row limit stops the step instead of writing an incomplete total, "Run only on certified definitions" holds back anything uncertified, and the run log names every definition and version the figures came from.
Import from dbt and OSI: Settings → Semantic objects → Import dbt / OSI… reads the semantic models and metrics of a dbt project, or an Apache Ossie (OSI) model, into the shared catalog or the project. Preview shows everything that would be created and everything left out with the reason — for example cumulative metrics, which have no equivalent yet. An id prefix keeps two imports apart, and nothing is written until you confirm.
Ask Greg builds on your metrics: Ask for a chart, a pivot or a KPI of a figure your organisation has defined and Ask Greg binds it to the metric itself, saying which definitions it rests on and whether each is certified. It can also propose a pipeline with a Governed metrics step, and a data story — choosing what to narrate, never the sentences. Every metric it names is checked against the catalog before you see the card.
MCP servers in Ask Greg: Connect MCP servers — ready-made templates for BigQuery, PostgreSQL, Jira and Confluence, or your own — in Settings → MCP servers & API. Tools are read-only unless you allow more, secrets stay in the encrypted vault, and Ask Greg uses a server only when your data policy and the server both allow it, with its results masked like any other data.

Version 1.6.0: Better every day

Dashboards open instantly when you come back: A dashboard you have opened before shows every chart, pivot, KPI and slicer straight away from its last answer and refreshes in the background — before, every visit started again from "Loading…". A small spinning mark in an object's title strip shows it is refreshing; PDF reports always fetch fresh numbers.
Faster sheet switching: Switching sheets no longer waits on the function reference and manual window while it is closed, and editing a cell no longer redraws the ribbon.
Charts read numbers safely on every database: Histograms, forecasts, cluster charts and range filters now treat a text value such as "n/a" as missing on every database. Before, the same chart failed on PostgreSQL, Redshift and Oracle — and on MySQL it quietly counted as 0.
Semantic filters only reach the widgets they belong to: A dashboard filter bound to one dataset no longer breaks widgets that read another one; the widget notes say which filters were skipped.

Version 1.6.0: Security & privacy

AI writes the words, never the numbers: In data stories, explanation boxes and alert e-mails, every figure is checked in code against the data before it is shown or sent, the AI sees only what your data policy allows, and AI-written text is marked as such, as the EU AI Act asks.
Imported models cannot carry unsafe SQL: A dbt or OSI import refuses any expression that could change data, read files or reach the network, rejects oversized or malformed files, and only lets you point the imported definitions at a connection you are allowed to use.
Less information to anonymous health checks: The public readiness check now answers only whether the app is ready; memory and queue details need the API token (ATX-SEC-004).
Cleaner installers: Third-party tooling files — one of which carried an upstream project's public coverage token — are no longer packaged, so a binary secret scan comes back clean (ATX-SEC-008).

Version 1.6.0: For IT & administrators

Public REST API: A versioned, read-only API (/api/v1) for project tables and SQL, saved connections, the semantic layer and dashboards, described in OpenAPI 3.1 with a reference page. It uses the same sign-in, roles, sharing, row- and column-level security, licence and audit as the app, with rate limits, paging, CSV and NDJSON output, and standard error documents.
Headless metrics: Other tools can list and query your governed metrics over the API and get exactly the numbers the dashboards show, with the SQL that produced them.
ATX AnalyticsTable as an MCP server: Let Claude, Copilot Studio, Cursor or your own agent find, describe and query governed metrics over the Model Context Protocol — read-only, with the viewer role, never writing SQL and never seeing a credential. Off until an administrator switches it on.
Publish gate and Restore: Optionally serve reports, agents, METRIC() cells and embedded dashboards only the version of each shared metric a steward certified, while authors keep editing. Any earlier version of a definition can be restored from its history in one click.
Embedded analytics: Show a dashboard inside your own web application with short-lived signed tokens your backend issues for one dashboard and one origin, locked parameters for per-customer row scoping, a small JavaScript SDK and revocation. Off until an administrator switches it on.
Licence component for AI narratives: Data stories, AI explanation boxes and AI paragraphs in alerts are a licence component of their own. Licences issued before 3 November 2026 include it automatically. The routes are audited, and the managed AI policy can switch them off with the rest of Ask Greg.
Require multi-factor sign-in: An OpenID Connect or SAML sign-in can be refused unless the identity provider confirms a second factor. Refusals are recorded in the audit trail. Off by default.
New Group Policy and Intune settings: Require multi-factor sign-in, require encryption of new project databases, and set how long the audit trail is kept — in English, German and Polish policy templates.
Enterprise Security pack: Single sign-on, writing the AI policy from the app, Mask PII, audit-trail review and bulk credential export are now a licence component of their own. Licences issued before 30 October 2026 include it automatically, so upgrading takes nothing away. The vault, encryption in transit, masking of what is sent to an AI and recording the audit trail are never gated.
Compliance evidence: A control matrix maps the product's controls to SOC 2 and ISO/IEC 27001 (no certification is claimed), and each installer now comes with signed build provenance.

Version 1.5.0 — 29 October 2026 — Control who sees what — down to the row — and sign in with your own identity provider

The enterprise access release. You can now share a single project, dashboard or connection with exactly the people who need it, and decide which rows and columns of your data each person sees. Sign-in works with any OpenID Connect or SAML identity provider, with users provisioned and removed automatically. And ATX AnalyticsTable connects to Snowflake, Databricks, Redshift, MySQL and Microsoft Fabric.
Row- and column-level security: One policy decides which rows each person sees and whether a column is shown, masked, pseudonymised or hidden — enforced on the server, on every path.
Share one object, not a whole role: Give a person or a group access to one project folder, dashboard, connection or mail profile — and hide it from everyone else.
Single sign-on with any provider: Okta, Auth0, Keycloak, Ping, OneLogin, Google, ADFS and any SAML 2.0 provider, next to Entra ID — with SCIM provisioning and instant revocation.
Five new data warehouses: Snowflake, Databricks SQL, Amazon Redshift, MySQL / MariaDB and Microsoft Fabric / Azure Synapse — no Java needed.

Version 1.5.0: good to know before you upgrade

1.5.0 installs over 1.4.0 and keeps all your projects, settings and licence. Sharing and data-access policies are off until someone sets them up, so everyone keeps the access they have today.
If you have scripts or integrations that call the server directly, check that they use the documented lower-case addresses: a different spelling (capital letters or a trailing slash) is now refused.
The new warehouse connectors are a preview: test each one against a non-production workspace before you roll it out.

Version 1.5.0: New

Row- and column-level security: Your organisation writes one data-access policy for a project, and every person sees only the rows meant for them — "the regions this person covers", say. Each column can be shown in full, masked, replaced by a stable pseudonym, left empty or hidden altogether. The rules follow the same people, groups and roles your sign-in already knows, and they are enforced by the server everywhere the data can go: queries, sheets, dashboards, exports and Ask Greg. Settings → Data access shows anyone exactly what they can see.
Share a single project, dashboard or connection: Click Access on a saved connection, a mail profile or a dashboard's header — or use Settings → Access for a project folder — and choose who may view, edit or manage it. Access to a folder covers everything inside it, grants can expire, and a restricted object simply disappears for everyone else: from the lists, the folder browser and Ask Greg. Nothing changes until someone restricts something.
Snowflake, Databricks, Redshift, MySQL and Fabric (preview): Five new connection types in Settings → Connections, each with the sign-in methods enterprises actually use — key pair, OAuth, managed identity, IAM and Entra ID single sign-on. Paste the JDBC URL your DBA sent you and the form fills itself in; it also shows the credential-free URL to hand back for a firewall request. They work everywhere a database does: Get Data, the query builder, pivots, exports, the scheduler and the SQL formula functions. These connectors are new — please try them on a test workspace before relying on them in production.
Join tables in the visual query builder: Add several tables to one query and join them by dragging a column onto a column of another table — inner, left, right or full joins, on one key or several. The tables can even come from different databases.
Choose which widgets a dashboard filter drives: A filter can now reach every matching widget, only the ones you tick, or all except them — and you can tell it which column to filter on in each widget when the names differ.

Version 1.5.0: Security & privacy

A high-severity access-check bypass is closed: Our security assessment found that unusual spellings of a web address — capital letters or a trailing slash — could slip past the server's access checks. Every address is now normalised before any check runs. The flaw was found and fixed before release, and the attack is re-run automatically on every build.
Up-to-date components: Third-party components with published advisories have been raised to their fixed versions, including the ones the new warehouse drivers brought in.

Version 1.5.0: For IT & administrators

Single sign-on with OpenID Connect and SAML 2.0: Set up Okta, Auth0, Keycloak, Ping, OneLogin, Google, ADFS, Shibboleth or any other provider in Settings → Identity → Single sign-on, alongside or instead of Entra ID. Sign-in is handled by the server, signatures and audiences are checked strictly, and unsolicited SAML responses are refused.
SCIM provisioning and instant revocation: Let your identity provider create, update and deactivate users and groups automatically. Deactivating someone ends their sessions at once, and an administrator can revoke every token issued so far in one step. Idle and maximum session limits are configurable.
Access reviews: Settings → Access shows administrators every grant, exports an access review as CSV, and answers "can this person open this?" with the reason. Every access decision, and every time an administrator reaches into a restricted object, is in the audit trail.
Built for a bank's software review: The warehouse connectors use each vendor's own driver — no Java runtime to install or patch — enforce TLS 1.2 or later, and keep every credential in the encrypted vault, never in a project file or log.

Version 1.4.0 — 25 October 2026 — A new name, clearly labelled AI and an urgent security fix

A short, focused release. Everything the AI writes is now marked wherever it goes — in the grid, in Excel and in PDF reports — so you and your readers always know which values a person checked. The desktop app is also protected against a newly published Node.js vulnerability, and the product takes its full name: ATX AnalyticsTable.
You always know what the AI wrote: AI answers are marked in the grid, and the mark travels with your data into Excel workbooks and PDF reports.
Protected against the Node.js Brotli flaw: A newly published vulnerability (CVE-2026-111497) is blocked in the desktop app, ahead of the official fix.
Same app, new name: AnalyticsTable is now ATX AnalyticsTable. Upgrading keeps your projects, settings and licence exactly where they are.

Version 1.4.0: good to know before you upgrade

Nothing to prepare. 1.4.0 installs over 1.3.0 and keeps all your projects, settings and licence.

Version 1.4.0: New

AI answers are marked — everywhere they go: Every cell filled by an AI formula, and every value a pipeline's AI step writes, carries a visible mark in the grid. The mark goes with the data: into Excel workbooks (as a cell note and in the document properties) and into PDF reports, so whoever reads your report can see which numbers and texts an AI produced.
Review AI answers before you rely on them: Answers written by a pipeline that ran on its own stay marked as unreviewed until a person has looked at them. Select the cells, right-click and choose Mark AI answers as reviewed.
Tell us when the AI gets it wrong: Give Ask Greg's answers a thumbs up or down, and use Report wrong AI answer on any AI cell. Administrators can export the feedback as CSV from the audit dialog.
A clear "you are talking to an AI" notice: Ask Greg now always shows that it is an AI assistant whose answers can be wrong, right under the chat box.
A safeguard for personal data: When an AI answer would be about one particular person's personal data, ATX AnalyticsTable notices and warns you. Your organisation can choose to keep the warning (the default), refuse such answers, or allow them.
Built for the new AI rules: Together these controls are designed to support the transparency, record-keeping and human-oversight duties of the EU AI Act (Articles 50, 12, 14 and 26) and the Colorado and California rules on automated decisions.

Version 1.4.0: Better every day

The product is now called ATX AnalyticsTable: You will see the new name in window titles, the start screen, shortcuts, the installer, the manual, the legal notices and all six translations. Nothing you rely on moves: the install folder, your settings, projects, licence, saved passwords and Group Policy settings all stay where they are, and .atx project files are unchanged.

Version 1.4.0: Security & privacy

Protection against the Node.js Brotli vulnerability: A flaw in how Node.js unpacks Brotli-compressed data (CVE-2026-111497) could let a malicious server run code on your computer. The fixed Node.js is not yet available in the desktop framework ATX AnalyticsTable is built on, so until it is, the app no longer asks servers for Brotli-compressed responses, and a server that sends one anyway gets a clear error instead. Ordinary servers are not affected, and the safeguard switches itself off once the fixed Node.js arrives.
Latest browser-engine security fixes: The desktop app now runs on Electron 44.9.3 (Chromium 152.0.7977.167), with every security fix released since the previous version. Further Node.js fixes for certificate checks will follow in the next update that carries them.

Version 1.4.0: Fixed

Kubernetes chart version: The Helm chart for server installations reported 1.2.0; it now follows the product version again.

Version 1.3.0 — 18 October 2026 — Forecasts, histograms, hyperlinks — and jobs that run as long as they need

The biggest release so far. Two new charts let you look ahead (forecasting) and look inside your numbers (histograms). Cells can link to the web, to other sheets and to other projects. Long queries and pipelines no longer stop after two minutes, sheets feel faster, and the manual now speaks every language the app does. Organisations get roles, an audit trail, Group Policy and Intune support.
Forecast chart: Project any value over time months ahead, with ranges you can trust and what-if scenarios.
Histogram chart: See how your values are spread — with density curves and Cp / Cpk for quality control.
Jobs that run for hours: Queries, pipelines and scheduled refreshes are no longer cut off at two minutes — and Stop really stops.
Hyperlinks in cells: Link to a website, an e-mail address, another sheet or another project, and jump back with Alt+Left.
Range filter for dashboards: Two sliders — from and to — filter every widget on a dashboard to a band of values.
The manual in seven languages: English, Polish, German, Spanish, French, Portuguese and Chinese — every section, function and shortcut.

Version 1.3.0: good to know before you upgrade

Projects you receive from other people now open paused. Choose Trust this project once, and they run as before.
Queries no longer stop after two minutes. If you relied on that as a safety net, set your own limit in Settings → Advanced configuration → Query timeout.
The installer now shows the End User License Agreement, which you accept to continue.
Running your own licence portal? Your own signing keys now need to be vouched for by ATXInstitute — contact us before upgrading.
Server installations on terminal servers or Citrix: users start ATX AnalyticsTable from the shortcut; a bookmarked local address no longer works on its own.
Installed a 1.x MSI per user? Remove it before installing the new per-machine MSI.
Moving a trial clock backwards no longer extends a trial.

Version 1.3.0: New

Forecast chart: Pick a date column, a value and a period (day, week, month, quarter or year), say how far ahead to look, and the chart draws the forecast with 80 % and 95 % prediction ranges. Leave the method on Automatic and ATX AnalyticsTable tests every method on your own history and keeps the most accurate one — or choose from ten, from a simple moving average to Holt-Winters. Add what-if scenarios (a flat percentage or compounded growth), a management adjustment, or one forecast per region or product with Other and Total. The forecast horizon can come from a cell, and an accuracy band shows how well the method did on periods it had not seen. Find it under Dashboard → Forecast or Reports → Forecast. Ask Greg can propose one for you.
Histogram chart: Show how the values of one numeric column are spread. Bins are chosen automatically, or by one of eight rules, a fixed count or width, or your own edges. Show counts, percentages, density or the sum of another column; split by a category side by side, stacked or overlaid; add a normal or density curve, a cumulative-% line and markers for mean, median, ±1σ and percentiles. Enter specification limits and you get Cp, Cpk and the share outside spec — ready for quality control. Click a bar and the whole dashboard filters to that range. On very large tables the counting happens in your database. Find it under Reports → Histogram or Dashboard → Histogram.
Range filter on dashboards: Dashboard → Range filter adds from / to fields with sliders over a number column. Every widget showing that column follows, and histograms count only the rows inside the range. It applies when you let go of the slider, so the dashboard does not flicker while you drag.
Hyperlinks in cells: Press Ctrl+K (or right-click → Insert hyperlink…) to link a cell to a web or e-mail address, a place on the same sheet, another sheet — even after it is renamed — or another ATX project, which keeps working when the folder moves. Follow a link with the small ↗ or Ctrl+Enter, and go back with Back or Alt+Left. Links move with their cell when you sort or insert rows, a link manager lists every link in the project, and Excel exports carry real Excel hyperlinks. Before opening a web link, the app shows you the real address it leads to.
The user manual in every language the app speaks: German, Spanish, French, Portuguese and Chinese join English and Polish — every section, function and shortcut is translated.
Backup & restore in one file: Settings → Backup & restore saves your settings, connections and saved credentials into one passphrase-protected file. Restore it on a new laptop or a fresh virtual desktop, with a preview of what will come back.

Version 1.3.0: Better every day

Long-running jobs: Queries, pipelines and scheduled refreshes can now run for up to two hours by default — before, every query ended at two minutes whatever the settings said. You can change both limits, or remove them, in Settings → Advanced configuration (Query timeout and Background job timeout). Stopping a job in the Query Monitor now stops the query on the database too, straight away.
Faster sheets: Charts and pivots come back instantly when you return to a sheet, showing their last answer while they refresh in the background (scheduled PDFs always use fresh numbers). Scrolling is smoother, and clicking a sheet tab responds at once.
Numbers in your own format: Values such as "1,5", "1 234,56" or "1.234,56" are now counted in SUM, AVERAGE, COUNT, SUMIF, VALUE and comparisons exactly as the grid shows them. New CSV exports use ";" as the separator where your decimal separator is a comma, so Excel opens them correctly.
Everything in your language: The first start picks your operating system's language when it is one of the six the app ships with. Status lines, error messages, pipeline results, dates, month and weekday names and the first day of the week all follow the interface language, and pop-ups are the app's own instead of the browser's.
PDF reports in Chinese and Japanese: PDF reports now print Chinese and Japanese text, and their own labels, row counts and dates follow the interface language.
Typing in Chinese and Japanese: Input methods (IME) can start a cell edit, and pressing Enter or Escape to pick a candidate no longer commits or cancels the cell — in the grid, the formula bar, Find, Ask Greg and other text boxes.
The engine rests when you do: After 24 hours with nothing happening, the database engine stops to free memory and file locks. A banner tells you, and your next action starts it again — no restart. Running queries and scheduled jobs keep it awake. Autosave no longer saves a project that has not changed.
Projects know which version saved them: Opening a project saved by a newer version of ATX AnalyticsTable warns you and offers to open it read-only, and anything an older version does not understand is kept when it saves.
A tidier SQL ribbon: SQL → Library opens the saved-query library; unfinished buttons are hidden until they are ready.

Version 1.3.0: Security & privacy

Projects from elsewhere open paused: A project you received from someone else no longer runs anything by itself — no schedules, refreshes, imports, exports or e-mails — until you choose Trust this project. Projects you created or saved yourself are trusted automatically.
Saving can no longer damage a project: A save is written to a temporary file first and then swapped in, and the previous version is kept as <name>.atx.bak. Saving into protected system folders is refused.
Ask Greg says it is an AI: A line under the chat and on its empty screen says that Ask Greg is an AI assistant whose answers can be wrong. AI formula cells show "Generated by AI" when you hover over them.
Cloud storage keys are kept in the vault: Keys for Azure Blob Storage, Google Cloud Storage and Amazon S3 are no longer stored in the browser, in project files or in settings exports. Existing keys move into the secure vault automatically the first time you start 1.3.0.
A broad security review: From our security audit of 11 October: limits on what SQL Ask Greg may run, protection against formula injection in CSV exports, stricter path checks, requests checked for sign-in before anything else, and backups checked before they are restored.
A hardened desktop app: The desktop app runs on Electron 44 (Chromium 152) with its files integrity-checked on Windows. Browser permissions are off by default (apart from the clipboard) and the window can only show ATX AnalyticsTable itself. Open and save dialogs remember the last folder again.

Version 1.3.0: For IT & administrators

Roles in server mode: Viewer, Analyst, Admin: Viewers open projects and dashboards and run read-only queries; Analysts can do everything a desktop user can; Admins manage credentials, shared connections, policies, the licence and the audit log. Roles come from Microsoft Entra ID app roles or groups. The desktop app is unchanged.
A tamper-evident audit trail: An append-only log records who did what — sign-ins, credential use, queries, exports, e-mails, backups, licence changes, AI requests and refused actions. Check that nothing was altered in Settings → Audit log, and forward it to your SIEM over syslog.
Group Policy and Intune: The MSI installs per machine and accepts settings at install time. ADMX templates let you turn off AI, cloud storage, e-mail, Git, the update check or individual connectors, and limit where exports may be written; users see "Turned off by your administrator".
Works on corporate networks: The app trusts the Windows certificate store, so AI, licence checks, cloud storage and REST calls work behind TLS-inspecting proxies. It follows Windows proxy settings, PAC and WPAD, and you can set a minimum TLS version for PostgreSQL, SQL Server and Oracle.
Resource limits: Per-query timeouts, a memory budget that queues work instead of running out of memory, optional engine worker processes and a memory cap for the backend on Windows.
Terminal servers and Citrix: The server installer gives every Windows user their own backend and profile, opened through a one-time launch link. Users start from the shortcut.
Monitoring and support: Prometheus metrics, a request ID on every response and log line, a redacted diagnostics bundle under Settings → About & support, local crash dumps, and a graceful shutdown.
Docker and Kubernetes out of the box: The container image and Helm chart start with working defaults and refuse configurations that cannot work.
Signed, scanned releases: Every release passes automated tests and a vulnerability scan before it is published, and ships with a software bill of materials (SBOM). Windows installers are code-signed and timestamped.

Version 1.3.0: Fixed

Negative currency: Negative amounts on dashboards read "-€3.3M" instead of "€-3.3M".
Merged cells: Inserting or deleting rows and columns now moves, grows or shrinks merged cells correctly.
Imported blocks with a header row: Disconnecting such a block no longer shifts the data up a row, and its last row can be selected and sorted again.
Linux shutdown: Engine workers now exit within about three seconds when the server stops, instead of running on and holding database files.
Apps & features: The server installer showed version 1.0.0 in Windows Apps & features; it now shows the real version.
Stability: An unexpected internal error can no longer bring down the backend.

Version 1.2.0 — 10 October 2026 — Ask Greg builds whole dashboards — and one click removes a filter

Ask Greg, the built-in AI analyst, can now put together a complete dashboard for you — widgets, filters, titles, annotations and timelines — and checks every piece against your real data before offering it. Removing a sheet filter is now one click, and scheduled jobs run at the right time on the days the clocks change.
Ask Greg builds dashboards: Describe what you want to see and get a whole dashboard of up to 20 widgets, ready to place with one click.
Remove filter, in one click: Transform → Remove filter takes every filter off the sheet and brings all hidden rows back.
Schedules that respect the clock change: Jobs set for a time that is skipped or repeated when clocks change now run once, at the right moment.

Version 1.2.0: good to know before you upgrade

Nothing to prepare. 1.2.0 installs over 1.1.0 and keeps all your projects, settings and licence.
A job scheduled inside the hour skipped in spring now runs just after the gap, not an hour early.

Version 1.2.0: New

Ask Greg builds dashboards and report objects: Ask Greg can now propose a dropdown filter, a title, annotation shapes (lines, arrows, rectangles, text boxes, callouts and more), a timeline — from milestones you type or a column of dates — or a whole dashboard of up to 20 widgets at once. Everything is checked before it is offered: a filter against the values its column really holds, a dashboard widget by widget. Placing it is a single undo step, and Remove takes a dashboard back as long as you have not changed it. Dashboards, filters and titles need the Dashboard module on your licence.
Transform → Remove filter: One button in Sort & Filter takes the whole filter off a sheet: every column filter is dropped, the rows it hid come back and the funnel icons disappear. Before, you had to click Filter again on each selection. Clear works as before and its tooltip now points to the new button; rows you hid by hand stay hidden.

Version 1.2.0: Better every day

Ask Greg opens its chat straight away: Clicking the Ask Greg tab now opens the chat panel on the right too. The tab is labelled "Ask Greg (beta)" in all six languages, to show the feature is still being refined.
Schedules and pipeline steps from Ask Greg, thoroughly tested: Every schedule Ask Greg can propose — for pipelines, alert rules, saved and pinned queries and PDF reports, at every kind of interval and across time zones — and every one of the 19 pipeline step types is now covered by automated tests, from the proposal right through to running it and reopening the project.

Version 1.2.0: Fixed

Several objects for "a new dashboard": When Ask Greg placed several objects on a new dashboard, each one created its own dashboard. They now share one, and new objects stack under what is already there instead of landing on top of it.
Scheduled jobs on the days the clocks change: A job set for a time that does not exist on the spring-forward day (for example 02:30 on the last Sunday of March in Poland) ran an hour early. It now runs at the first valid moment after the gap in every time zone, and a time that happens twice on the autumn fall-back day runs once.

Version 1.1.0 — 26 September 2026 — AI answers row by row, an import history, and schedules that retry

A release about everyday reliability. The AI step in a pipeline can now answer once per row, so a column of comments becomes a column of classifications. Every import is recorded in a history you can check, scheduled runs can retry and tell someone when they fail, and Find shows its progress with a Stop button. Get Data is reorganised so every database, file type and cloud store has its own button.
AI answers, one per row: Point the AI step at a column of cells and it answers the question for every input row — classify comments, tidy names, tag records.
Import history: Get Data → History shows every import: what was read, from where, how many rows and how it ended.
Schedules that retry: A scheduled run can try again up to five times, and e-mail someone if it still fails.
Find you can watch — and stop: Searching big sheets shows progress, matches found so far and a Stop button, and the window stays responsive.

Version 1.1.0: good to know before you upgrade

Nothing to prepare. 1.1.0 installs over 1.0.0 and keeps all your projects, settings and licence.
Whole-sheet exports may come out narrower than before: empty trailing columns and rows are no longer written.

Version 1.1.0: New

AI analysis answers row by row: In a pipeline's AI analysis step, the size of the destination you pick decides what it does. One cell gets a summary, as before. A column of cells asks the question once for every input row and writes the answers down the column, row for row — how a column of free-text comments becomes a column of categories. The step never writes past the cells you chose, rows the AI did not answer are left empty rather than showing yesterday's answer, and several rows go to the AI in one call to keep it fast and inexpensive. If your organisation's AI policy forbids sending individual rows, the step says so instead of guessing.
Get Data, sorted by what the source is: Three groups replace the old "Data Formats": Databases (BigQuery, SQL Server, PostgreSQL, Oracle, DuckDB, SQLite), Files (Excel, Text/CSV, JSON, NDJSON, Parquet) and Cloud Storage (Google Cloud Storage, Azure Blob, Amazon S3). Every engine now has its own button — before, five of them were only reachable through the Oracle dialog. Clicking one preselects a connection of that kind, or tells you there is none yet.
Scheduled imports from cloud storage: Schedule… in the bucket browser puts a cloud file on a schedule. A name such as sales_${asOfDate}.parquet is worked out at each run, so every run reads that day's file.
Get Data → History: Every import records what was read, from where, onto which sheet and cell, how many rows, how long it took, who was signed in and how it ended. The last 200 are kept with the project and saved as each run finishes, so an unattended run's result is there even if nobody pressed Save. "Nothing new to read" shows as Skipped rather than as a success, so a stalled feed no longer looks healthy.
Retries and failure e-mails for schedules: A schedule can retry up to five times with a pause between attempts, and send one e-mail after the last attempt fails — a run that succeeds on its second try sends nothing. A run that timed out is never retried, because the work may still be running and a second attempt could load the same rows twice.
File → Close Project: Close the current project and keep the app open on an empty one, instead of quitting and starting again. Unsaved changes get the usual Save & Close / Close without Saving / Cancel question. The project's files are released straight away, so you can move, rename or zip the folder at once.

Version 1.1.0: Better every day

Find shows what it is doing: Searching across several grids no longer freezes the window. The Find dialog shows which grid it is on, how far it has got, how many matches it has found, a progress bar and a Stop button that keeps what was found so far. Very large searches stop at 20,000 matches and say so.
The SQL tab opens your saved queries: Clicking SQL now opens the Queries panel in the sidebar, one click instead of two. Prefer the sidebar to stay as you left it? Turn off Settings → Query behaviour → Open Queries with the SQL tab.
Exports stop where your data stops: Exporting a whole sheet no longer adds empty columns and rows from the edges of the grid. Blank columns between your blocks, or a report that starts further in, are kept exactly as you laid them out, and a range you chose yourself is never trimmed. The export dialog now shows the width beside the row count.
Clearer feedback in connection settings: "Saved" with the time appears right beside the Add/Update button, and BigQuery's Save explains why nothing happened when the key box is empty.

Version 1.1.0: Fixed

Ask Greg with GPT-5 on Azure AI Foundry: Ask Greg now works with GPT-5 and o-series deployments on Azure AI Foundry. Whichever endpoint address you copy from the Foundry portal is understood, a model version typed as the API version no longer fails, and a remaining error says which three fields to check.
New Project really starts fresh: A new project used to inherit the previous one's saved queries, schedules, pipelines, filters, reports, column widths and more — and its schedules kept running. Now a new project starts empty and opens on its blank sheet. Your connections stay, because they belong to you, not to a project.
Test Connection straight after Update: Updating a connection used to clear the form, so you could not test it without reopening it. The form now stays loaded after Update.
No more swallowed clicks in Settings: After reading a connection test result, the next click in Settings used to only close the result window. The result window no longer blocks the rest of Settings.
Search in: Formulas: Find could never match anything when searching formulas. It now finds them.

Version 1.0.0 — 18 September 2026 — The first general-availability release

ATX AnalyticsTable 1.0 is the first release for general use. On top of everything in the preview, it brings Excel-style IF, AND, OR, NOT, ISBLANK and IFERROR to the grid, makes SQL formulas give the same answer whether they run in the sheet or in your database, and makes every error message easy to copy into a support request. The installers now carry ATXInstitute's legal notice.
IF, AND, OR, NOT, ISBLANK, IFERROR: The everyday decision functions you know from Excel now work in the grid.
Same answer in the sheet and the database: Text formulas give the same result in the grid and when pushed down to DuckDB, BigQuery, PostgreSQL, SQL Server or Oracle.
Error messages you can copy: Select, right-click or use the copy button on any error in a pivot, chart, KPI or report object.

Version 1.0.0: good to know before you upgrade

Text functions now see a cell's real value. An empty cell is empty (=LEN of it is 0, not 1), leading zeros such as 007 are kept, and spaces around imported values are no longer trimmed away — so =LEN may report longer values on data from fixed-width files. To get the old result, wrap the reference in TRIM: =LEN(TRIM(C1)).
SQL text formulas on a blank cell now answer as Excel does: LEN_SQL gives 0 and CONCAT_SQL keeps the other text, instead of an empty result. VALUE_SQL of text that is not a number gives 0.
Emoji and accents written as two characters can still count differently in the grid and in a database; the database's answer is the more accurate one.

Version 1.0.0: New

Conditional functions: IF, AND, OR, NOT, ISBLANK and IFERROR now work in the grid. IF without an "else" leaves the cell empty instead of writing FALSE, so a column of numbers stays clean. ISBLANK treats zero as a value, not as blank. IFERROR catches every error the app can show — for example =IFERROR(UPPER_SQL(C3), UPPER(C3)) falls back to the grid when a database is unreachable.
BigQuery tables that require a partition filter: Such tables used to fail with a bare BigQuery error. They now open on their newest partition, and the preview says which one, so nobody mistakes a slice for the whole table.
A recovery panel instead of a blank window: If something in the grid crashes, you get a panel with a Copy details button instead of an empty page, and server errors are written to the log file you attach to a bug report.
Legal notices in the installers: The installers, the user manual PDF and the server image now carry ATXInstitute's legal notice.

Version 1.0.0: Better every day

SQL formulas agree with the grid: The *_SQL formula family now behaves the same in the sheet and in DuckDB, BigQuery, PostgreSQL, SQL Server and Oracle — across more than 650 test cases per database. A blank cell is treated as empty text rather than "no answer", VALUE_SQL reads "1,234.50" as 1234.5, and TRIM_SQL collapses inner spaces like Excel does.
Errors you can copy: Messages inside pivots, charts, KPIs, slicers and report objects can be selected, copied with Ctrl+C, copied from the right-click menu or with a small copy button. In the desktop app, right-clicking selected text anywhere offers Copy, Cut, Paste and Select all.
Correct values from BigQuery, PostgreSQL and DuckDB: Dates, timestamps and exact numbers arrive as real dates and numbers, so they sort and add up correctly. PostgreSQL NUMERIC values keep every digit.

Version 1.0.0: Fixed

Sheet-range pivots and charts in the installed app: Pivots, charts, heatmaps and KPIs built on a range of a sheet failed in the installed app with an "unsafe-eval" message. They work again.
BigQuery tables with no columns: Every BigQuery table opened empty because the column list was read from the wrong place.
Text in BigQuery with line breaks or apostrophes: Text functions failed on cells containing a line break (a pasted address, a note), and a value such as o'neill could produce a wrong result. Both are fixed.
Result sets capped by mistake: Some queries had their row limit applied twice, which quietly cut results short — and on BigQuery billed for the larger scan.
Switching a table's source: Moving a table from the local database to a connection no longer crashes the window.

Glossary terms on this page

ATX Institute on LinkedIn

This site needs JavaScript for the full experience. Every page above is reachable without it.