ATX AnalyticsTable is a local analytics client for PostgreSQL: the query runs on your server, the result is copied into a DuckDB database inside the project for fast columnar analysis, and tax IDs, e-mails and names are masked before the extract is shared. It connects to self-hosted PostgreSQL and to the managed services on Azure, Google Cloud and AWS.
The PostgreSQL connection form in ATX AnalyticsTable 1.7, filled in with the example values below. Secrets go into the encrypted vault, never into the project.
Map the managed identity to a read-only role (as the server's Entra admin): SELECT * FROM pgaadauth_create_principal('atx-reporting-mi', false, false); GRANT CONNECT ON DATABASE finance TO "atx-reporting-mi"; GRANT USAGE ON SCHEMA public TO "atx-reporting-mi"; GRANT SELECT ON ALL TABLES IN SCHEMA public TO "atx-reporting-mi";
Reporting query — tax IDs hashed in PostgreSQL: SELECT date_trunc('month', i.issued_at)::date AS month, i.cost_center, encode(sha256(convert_to(lower(v.tax_id), 'UTF8')), 'hex') AS vendor_key, sum(i.amount_net) AS amount_net FROM invoices i JOIN vendors v ON v.vendor_id = i.vendor_id WHERE i.issued_at >= date_trunc('year', now()) GROUP BY 1, 2, 3 ORDER BY 1, 2; — sha256() is built in from PostgreSQL 11.
Then reshape locally — a Run SQL step on the project's DuckDB: -- "input" is the table the previous pipeline step produced (the import above). -- This runs on your machine, in the project's DuckDB, not on the database. SELECT region, SUM(revenue) AS revenue, COUNT(DISTINCT customer_key) AS customers FROM input GROUP BY region ORDER BY revenue DESC; — Column names follow the query above; on the local engine the pipeline's current table is registered as input before the script runs.
PostgreSQL row-level security and column grants apply to the role that runs the query. With Entra sign-in and on-behalf-of turned on, queries run as the signed-in person, so your RLS policies apply per user — and if the token exchange fails, the query fails rather than falling back to the application's identity.
Reading fresh rows needs a route to the server. The imported copy lives in the project's DuckDB on your disk, and analysis, pipelines and reports over it work offline; the licence validates offline and no ATX service sees the data.
Analytical queries — group-bys, pivots, window functions over millions of rows — are what a columnar engine is for. A sum over 100 million rows takes 0.35 seconds on a laptop in AnalyticsTable's published benchmark.
verify-full. require encrypts but does not check who answered; verify-full is the only mode that defeats a man-in-the-middle.
Yes, with a data-access policy on the imported tables: per user, group or role it filters rows and shows each column clear, hashed, masked, empty or not at all — enforced by the backend on every path.
This site needs JavaScript for the full experience. Every page above is reachable without it.