Launching today

Chronos Gateway
The firewall between AI agents and your production data
2 followers
The firewall between AI agents and your production data
2 followers
AI agents have already wiped production databases. Chronos Gateway is how you let yours write to real customer data anyway. Every agent gets a run token, never your Postgres password, and a private copy of one customer's rows. Safe changes merge back in one transaction, every row checked against live data. Deletes and big changes stop and wait for a human, who approves in plain English. Your row-level security still decides what it can write. Self-hosted. Free. Postgres 12-18, Supabase, Neon.






Free
Launch Team

Render Sandboxes: a computer for every agentRun untrusted code safely. Idle sandboxes cost nothing
Promoted

Hi Product Hunt, Abhishek here again. On Friday we launched Chronos DB. The first thing almost every team asks us is whether they'd have to move their data into a new database. They don't want to, so this is the part that sits beside the Postgres you already run.
Your backend opens a run for one task and one customer. The agent gets a run token, not your password, and works over SQL or MCP in a private copy of that customer's rows. Columns you mark are masked, so it reads a hash or the last four characters instead of the real value.
When the agent is done, your rules decide. By default, up to 500 changed rows and no deletes are written to Postgres on their own, in one transaction, with each row checked against what's live. If someone edited a row in the meantime, the change stops and their edit stays. Anything outside the rules waits for a person, who sees "Remove lead Hooli" instead of a SQL log and approves from the console, a one-time link, Slack, or a widget in your own app.
Your app keeps reading and writing Postgres exactly as it does today.
Where it stands: it's self-hosted only, licensed BUSL-1.1 (not open source), and tested on Postgres 12 to 18, Neon and Supabase. RDS and Cloud SQL aren't tested yet.
If your agents write to customer data today, what's the change you'd least want them to get wrong?