Launching today
Antelog
Stops AI coding agents from leaking or destroying your data
40 followers
Stops AI coding agents from leaking or destroying your data
40 followers
AI coding agents are amazing. Until one deletes your database, or leaks your secrets. Antelog stops the disasters before they happen, and shows you everything your AI did: every command, every website, every bit of data it sent. Works with Claude Code, Cursor and Codex. Free.
Launch Team
AWS Startup AdvisorPersonalized AWS guidance that meets you where you’re already building
Promoted
📌 Hey Product Hunt! I'm Yash, founder of Antelog.
In April, an AI agent deleted a company's entire production database in seconds. The backups went with it. The developer had only asked it to fix a small error.
That's when I realized: we're giving AI agents the keys to everything, with no brakes.
So I built Antelog.
🛑 It stops disasters. If your AI is about to:
• delete your database or backups
• wipe your files or overwrite your team's code
• show or send your passwords and keys
• send your data to a stranger because a web page tricked it
…Antelog blocks it, before it happens.
👀 It shows you everything. Every command your AI ran, every website it visited, every piece of data that left your computer. Written down, and impossible to secretly change.
🧠 Built on research. Antelog is built on LogAct, a research paper published by Meta, which records every AI action before it runs. We added the protection on top.
Works with Claude Code, Cursor, Codex and more. One install. Free.
What's the scariest thing your AI agent has ever tried to do? 👇
I'll be in the comments all day!
@yash_sarwaiya How does Antelog handle false positives without solwing down legitimate AI agent actions?
cool launch!! how often does it block stuff i actually wanted, like wiping a local test db on purpose
@igorblink
Thanks! Wiping a local test database won't be blocked. Antelog knows the difference: databases on your own computer with names like app_test or dev, and SQLite files, can be wiped freely. It only blocks wiping production data and backups.
We tested it on thousands of everyday commands, and it rarely blocks something you wanted. If it ever does, just run that command yourself. 🙂
what about a python script the agent writes that reads .env, caught too?
scariest one for us wasn't even a delete. an agent did git add -A and pushed another session's half done work straight to main lol. we ended up writing our own hook rules for stuff like that, can you add project specific rules in antelog or is the list fixed?
@digi_dinos
Haha, that one hurts.
Yes, you can add your own rules! One line per rule, in plain words:
antelog rules add "git push * main"
From then on, Antelog blocks the AI from pushing to main, and you still can. For teams, admins set company rules once on the dashboard, and they apply on every developer's computer.
@yash_sarwaiya oh nice, plain words is exactly what we needed. git push * main would've saved us that day lol
@digi_dinos :) .. yes, plain words are the way to go!!
@yash_sarwaiya haha yep. good luck w the launch today 🙌
does it catch stuff that goes thru mcp tools too or only shell commands? half the time our agent hits the db via an mcp server not psql so thats where id actually worry
@bacelj
Yes, MCP tools too! Antelog checks every tool call the agent makes, not just shell commands. So if your agent sends DROP TABLE or wipes a table through a db MCP server, it's blocked the same way as through psql. The same goes for MCP tools that delete cloud db or send data outside.
the april story where a small fix ended with prod and the backups gone is the exact thing that keeps people from giving agents real access
when Antelog blocks a command mid task, does the agent get told why so it can try a safer route??? or does the session jsut stop there???
@thamibenjelloun
Hey Thami, great question! Before your agent runs any command, the command goes to Antelog first. A small team of checkers looks at it and decides in a milliseconds: safe, so it runs, or dangerous, so it's blocked.
The agent keeps going with the rest of the task, just without that command. If it really needs it, it will ask you to run it yourself 🙂