Try it
Watch the gate decide. Then run it on your machine.
Below, thirteen real requests from the public demo are replayed: pick one and follow the statement through the gate. Then the command to start the public image, and the ways to get started.
Try it without installing anything
Ask the assistant for something. Watch PostgreSQL decide.
Thirteen real requests from the public demo: six everyday tasks and seven dangerous ones. Pick one and follow the statement through the gate: which check stops it, how many rows it would touch, and what is left in the database.
The request, in plain language
Pick a request to start.
The statement the model proposed
-
agent_gate.propose()the gate’s checks- kind_allowed
- no_writing_cte
- keeps_its_context
- no_amplification
-
agent_gate.commit()the real effect, measured by the databaserows touched — limit 5
The database afterwards, read by a superuser
A replay, in your browser, of the public demo’s deterministic run (--model stub): the same SQL and the same verdicts anyone gets by running it. Nothing executes here. With the real model (Qwen3-8B on CPU) the SQL changes and the safety result does not: with the database deciding, 0 of 7 trap tasks broke out.
Try it free
One command. PostgreSQL 19 with the gate and five more extensions.
The public image ghcr.io/latamflows/ygg-pg:0.3.6 ships PostgreSQL 19 RC2 on Red Hat UBI 9, with six of the seven extensions ready for CREATE EXTENSION (pg_recall_guard is not included) and pg_agent_gate already in shared_preload_libraries. It only starts with a password (POSTGRES_PASSWORD, SCRAM authentication) and runs as non-root. It is a preview for evaluation and must not be exposed to the internet. For production, see the Enterprise edition.
- Base
- PostgreSQL 19 RC2 · UBI 9
- User
- postgres, with the POSTGRES_PASSWORD password
- Port
- 5432
- Status
- Preview, for evaluation
Start the container
podman run -d -p 5432:5432 -e POSTGRES_PASSWORD=change-me ghcr.io/latamflows/ygg-pg:0.3.6 Connect (it asks for the password) and install the gate
psql -h localhost -U postgres -c 'CREATE EXTENSION pg_agent_gate' Reproduce it
CREATE EXTENSION pg_agent_gate;
CREATE TABLE orders (id int primary key, total numeric);
INSERT INTO orders SELECT g, g*1000 FROM generate_series(1,8) g;
CREATE ROLE assistant LOGIN PASSWORD 'assistant';
ALTER TABLE orders OWNER TO assistant;
SELECT agent_gate.register_agent('assistant', 'assistant', 'orders demo agent', p_max_rows => 5);
-- then: psql -h localhost -U assistant -d postgres and try: DELETE FROM orders; With this setup, as a superuser, we got the output shown on the first screen. Then connect as assistant and type DELETE FROM orders.
How to start
Three paths, depending on where you are.
None of them asks you to trust a promise: the first you do on your own; the other two start with a conversation and end with deliverables you can verify.
- 01
Evaluate today, for free
For your technical team, on their own machine.
The steps
- 1. Start the public image with one command
- 2. CREATE EXTENSION pg_agent_gate and register an agent with its limits
- 3. Connect your model (or the demo) and watch what it refuses and what it keeps
What you get
You see the gate working on sample data, without asking anyone for anything.
Start the image - 02
A pilot on your database
For production, with the Enterprise edition.
The steps
- 1. PostgreSQL 18 on Red Hat’s PostgreSQL image, with a user and a password
- 2. One agent of yours, with limits agreed with your team and a dry_run reviewed by a person
- 3. An auditable record of every proposal from day one
What you get
Scope, timeline and price are agreed with you. There is no public price yet.
Talk about a pilot - 03
Audit first
If you already have agents connected and do not know what they can touch.
The steps
- 1. Review of roles, row-level policies and extension surface
- 2. Every finding with the SQL that proves it, reproducible on your database
- 3. A prioritized closing plan, with or without the gate
What you get
You know exactly what an agent can do on your database today, before deciding anything.
Request an audit
Contact
Write to us. We will get back to you.
Enterprise edition, security audit or agent-to-database integration: one email with two lines about your case is enough. We answer in English and Spanish.
manu@latamflows.com A person reads the email. If you would rather write directly, the address is above.
The seven extensions, the image and the contributions to PostgreSQL core are signed by Manuel Reyes Bravo, from Chile. To talk code, repositories or the PostgreSQL lists, his personal email:
Personal email
manuelreyesbravo@gmail.com Also through GitHub issues: pg_agent_gate/issues