UnitAutogen for PostgreSQL

Your application is tested. The logic inside your database usually is not. UnitAutogen writes the tests for it.

Beta (v0.5.1-beta) · PostgreSQL 14-18 · Windows and Linux · Community Edition (free), with Team and Enterprise modules on top.

Your application has a test suite. The functions, procedures and triggers running inside your PostgreSQL database almost certainly do not, and it is not because your team is lazy. Testing database logic by hand means crafting seed data that forces each IF, CASE, EXISTS and exception path to run, one branch at a time. It is tedious enough that it rarely gets done, so the most critical logic you ship, the code closest to your data, goes out untested and breaks quietly: a wrong WHERE, a missing INSERT, a branch that never once executed.

UnitAutogen does that work for you. Point it at a database and it generates the tests, and the data that drives each branch, as standard pgTAP, runs them, and reports honest branch coverage. No test written by hand. We generate the tests, not just synthetic data: every generated test asserts a real, observed effect.

The guardrail for AI-written database code

AI assistants now write a lot of database logic, and they write it fast. That is exactly where a silent mistake costs the most, and where it is hardest to catch by eye.

Say you ask your AI editor to adjust the tiers in a membership_tier function. The edit reads fine, so you approve it, but it flipped one boundary:

-    ELSIF p_points >= 1000 THEN RETURN 'gold';
+    ELSIF p_points >  1000 THEN RETURN 'gold';

Now a customer with exactly 1000 points silently drops from gold to the tier below, the kind of thing no code review catches. But UnitAutogen already captured how that function behaved when it was correct, as standard pgTAP tests you committed, so the change fails one of them before it ships:

not ok 7 - membership_tier :: p_points = 1000
    #   have: silver
    #   want: gold

That is the guardrail. It pins today's behaviour as a test suite you did not have to write, so the next change, from an AI or a person, that moves that behaviour fails before it ships instead of in production. Because the tests are generated, it keeps pace with code produced faster than anyone could test it by hand. Integration testing extends the same net across the whole flow, functions calling functions and the triggers their writes fire.

Best of all, you can hand the whole loop to your AI assistant. Tell it once to use UnitAutogen as its guardrail, then just keep asking for features: it generates the tests, runs them on every change, and fixes what breaks before it ever shows you the code. See how to set that up, with the exact prompt and a real session, or the full worked example.

Unit testing is free and where most developers start. From there the tool climbs: integration testing across your functions; seccover, a full security-coverage pass that checks whether your row-level security and multi-tenant isolation actually hold and whether personal data leaks through error messages; and hipaa, which turns that into stamped HIPAA evidence. More on that below.

The one guarantee

UnitAutogen never gives you a fabricated, false-passing, or flaky test. If a branch cannot be tested reliably, the report tells you so, in plain language, rather than showing a green that would mislead you. So every report comes back with no failing and no false tests, and a visible, explained gap always beats a false pass. That is what makes the result safe to gate your CI on.

Who it is for

UnitAutogen serves the whole path from writing database code to signing off on it, and grows from a free developer tool into enterprise-grade assurance. What you can do, and who it is for:

Capability	Who it is for	Edition	What you get
Unit + trigger testing	Developers	Community (free)	A full pgTAP test suite plus branch/statement coverage for every function and trigger, generated and run for you.
Integration testing	Developers, QA	Team	Tests across your functions, run for real together, so interactions and cascades are exercised, not just units.
seccover	Developers, QA, tech leads / CIO, compliance	Enterprise	Proof that RLS and multi-tenant isolation are actually enforced across your functions, views and triggers, unfiltered.
PII audit (pii)	Compliance, developers	Enterprise	Finds personal data leaking through error and RAISE messages. Runs inside seccover by default.
HIPAA evidence (hipaa)	Compliance, auditors, tech leads	Enterprise	seccover plus an audit-log check, as one stamped report mapped to the HIPAA Security Rule.

The Community Edition, unit and trigger testing, is the everyday developer tool and is free forever. seccover is the one that platform leads, CIOs and compliance officers care about, and the one that carries the most value, so it gets its own section next.

seccover: prove your database actually isolates tenants

Most PostgreSQL apps that hold more than one customer's data rely on row-level security and careful function logic to keep tenant A from ever seeing tenant B. The problem is that nobody can easily prove it stays true as the code keeps shipping. A single new function, a SECURITY DEFINER helper, a policy that was loosened for a bugfix, and the isolation quietly breaks. You find out in an incident, or an audit, not in a test.

seccover is built for exactly that. It exercises your functions, views, triggers, RLS policies and column grants as different tenants and reports, per object, whether isolation is enforced or leaking, with the evidence and a ranked list of the database changes that close each gap. Its PII audit catches personal data escaping through error and RAISE messages, the leak nobody writes a test for. How each verdict is proven: How seccover proves its verdicts.

Two things make it procurement-grade rather than a one-off script. It is continuous: a full scan is meant to run on every deploy, because a database that redeploys weekly is a database whose isolation guarantees are a week stale. And it is self-hosted with no telemetry: nothing about your schema or data leaves your perimeter, which is usually the first question a security review asks. If you need auditable evidence that your data logic is safe, this is the module that produces it.

Compliance evidence

Because seccover emits standard pgTAP tests and dated, reproducible reports, its output doubles as evidence you can hand to an auditor. The isolation results map onto the controls that require you to prove data is reachable only by those authorized:

• HIPAA Security Rule: unitautogen hipaa runs seccover, adds an audit-log check, and writes one stamped report for the safeguards the database layer can evidence:

  • Access control (45 CFR § 164.312(a)(1)): electronic protected health information in a shared database is reachable only by the tenant it belongs to.
  • Audit controls (§ 164.312(b)): whether every data change is recorded, where, and with which values.
  • Integrity (§ 164.312(c)(1)): whether a client role can forge, alter or erase the audit record.
  • Authentication (§ 164.312(d)): what identity the row policies trust, and whether a client can claim another.
  • Evaluation (§ 164.308(a)(8)): run on every deploy, the dated reports show the controls were evaluated over time.

  The PII audit catches PHI escaping through error and RAISE messages. Encryption, transmission security, login verification and administrative safeguards are outside the database layer and stay with you. See HIPAA evidence.

• SOC 2 (Trust Services Criteria): logical access controls (CC6.1) and role-based, least-privilege access (CC6.3), plus the Confidentiality criteria (C1) where the data is sensitive. Because the scans run on every deploy, the reports show these controls operated across the audit period, not just that they were designed.

This is supporting evidence for your auditors, not a certification, and it stays inside your perimeter: seccover is self-hosted, so the proof is produced on your own infrastructure.

Compliance officers and platform leads: contact us for a seccover evaluation on your own schema.

Install

Quick install - downloads the newest release (pre-releases included), checks its SHA-256, and unpacks it:

• Linux (into ~/.local/bin):

  curl -fsSL https://raw.githubusercontent.com/unitautogen/unitautogen-postgresql/main/scripts/install.sh | bash
  

• Windows PowerShell (into %USERPROFILE%\unitautogen):

  irm https://raw.githubusercontent.com/unitautogen/unitautogen-postgresql/main/scripts/install.ps1 -OutFile install.ps1
  powershell -ExecutionPolicy Bypass -File .\install.ps1
  

Or by hand: download the build for your platform from the Releases page (the newest one may be marked Pre-release), check it against SHA256SUMS.txt from the same release, and unzip it.

• Windows: dist-Windows-<version>.zip contains unitautogen.exe. Put its folder on your PATH, or call it by full path.
• Linux: dist-Linux-<version>.zip contains the unitautogen binary. Run chmod +x unitautogen (a zip does not keep the exec bit) and put it on your PATH. Needs x86-64 with glibc 2.31 or newer (Debian 11+, Ubuntu 20.04+, RHEL / Rocky / AlmaLinux 8+); not arm64.

Same commands, flags and reports on both platforms. Check it runs with unitautogen --version.

Beta note (Windows): the beta binary is not yet code-signed, so Windows SmartScreen may show "Windows protected your PC" on first run. Choose "More info" then "Run anyway". To avoid the prompt, right-click the downloaded zip, choose Properties, tick Unblock, then extract it. The install script does this for you. Signing lands when the tool leaves beta.

Extension setup for every platform, including managed/cloud PostgreSQL, is in docs/prerequisites.md.

Prerequisites

• The unitautogen binary (above). One self-contained executable, no Python install.
• PostgreSQL 14-18, with two extensions installed in the database you are testing:
  • pgTAP - the framework the generated tests are written in.
  • plpgsql_check - the coverage meter.

Point it at a disposable copy of your database, never production: it seeds rows and runs real statements, each rolled back.

Both extensions are required. UnitAutogen checks for them first and stops with a clear message (exit code 3) naming the exact CREATE EXTENSION to run if either is missing, so you never get a half-finished run. It works with any PostgreSQL 14-18, self-hosted or managed, including Supabase, Neon, Amazon RDS and Aurora, and Azure Database for PostgreSQL. Setup for every platform is in docs/prerequisites.md.

Quick start

Generating unit tests is the default action, so there is no command word:

unitautogen --schema public --out unit_tests --html report.html --json report.json

You get one test_<routine>.sql per function in unit_tests/unit/public/functions/, behaviour tests for triggers in unit_tests/unit/public/triggers/, an HTML report, a JSON summary, and the CI report bundle in unit_tests/unit/. Run a generated suite yourself at any time:

psql -d mydb -f unit_tests/unit/public/functions/test_transfer_funds.sql
psql -d mydb -c "SELECT * FROM runtests('test_transfer_funds'::name);"

In a Supabase project, unitautogen --supabase --schema public writes the tests where supabase test db runs them. See Supabase mode.

Walkthrough with real output: docs/quickstart.md.

Documentation

Document	What it covers
Overview	What the tool is and the five commands.
Prerequisites	PostgreSQL version, pgTAP and plpgsql_check setup.
Connecting to your database	Connection options.
Quick start	First run, end to end.
The five commands	Unit, integration, seccover, pii and hipaa, in full.
Supabase mode	Running any command inside a Supabase project.
HIPAA evidence	The hipaa command and its report, safeguard by safeguard.
How seccover proves its verdicts	How leaks are detected, in plain language.
Reading the report	How to read coverage and verdicts.
Worked example	A passing test and a failing test, up close.
AI guardrail (Supabase)	Using it as a guardrail for AI-written code, step by step.
The config file	Add or correct the PHI, personal-data and audit tables it detects.
CI/CD integration	Wiring it into your pipeline.
Command quick-reference	Every command, option and exit code.
FAQ and troubleshooting	Common questions and fixes.

Editions and licensing

• Community Edition (free, forever): unit and trigger test generation, coverage, and CI exit-code gating. This is the everyday developer tool.
• Team: adds integration. For teams that need cross-function testing.
• Enterprise: adds seccover, pii and hipaa, plus support and licensed, unwatermarked, audit-ready output. This is the tier for regulated and multi-tenant products.

Editions and prices: unitautogen.com/postgresql/pricing. For Team or Enterprise licensing, contact licensing@unitautogen.com. Use of the Software is governed by the licence terms.

Privacy

UnitAutogen runs entirely on your side. There is no telemetry and no phone-home; nothing about your schema or data leaves your machine. Tests are generated against stand-in data with the shape of your real tables, so no production data is required to run it. This is what makes it safe to point at sensitive and regulated databases.

Status

Active beta (v0.5.1-beta). The engine has been validated at high coverage across a range of real schemas, but you will surface things on production schemas nobody has run it against yet. Bug reports are the most valuable thing you can give us right now.

Support

• Bug reports and feature requests: open an issue.
• Questions: email hello@unitautogen.com.
• Security or vulnerability reports: see SECURITY.md or email security@unitautogen.com privately.
• Team / Enterprise licensing: licensing@unitautogen.com.

Related

For Row-Level Security policies alone there is rlsautotest, a free, open-source (Apache-2.0) tool that generates a pgTAP suite proving, per table, command and identity, who can read and write which rows: unitautogen/rlsautotest (pip install rlsautotest). seccover runs the same RLS policy tests as one of its checks.

UnitAutogen also has a SQL Server edition (built on tSQLt): unitautogen/unitautogen-public-repo. One brand, two database platforms. Learn more at unitautogen.com.

---

© 2026 Munaf Ibrahim Khatri. All rights reserved. UnitAutogen™ is a trademark of Munaf Ibrahim Khatri.