For engineering leads and companies whose AI agents can reach real systems.
The problem
Your developers use AI coding agents. Those agents run on many laptops and servers, and many of them can reach production: databases, cloud accounts, git. One wrong command can delete a database and its backups in seconds.
You can’t watch every laptop. And when something goes wrong, you need to know what the agent did, who allowed it, and why.
How it works for a team
- Each person installs Antelog. It checks every action on their computer before it runs, and it is free.
- On the Agents page of the online dashboard, you add each laptop or server. You get a few setup commands to paste on that machine.
- From then on, the machine sends its history to the dashboard every 5 minutes. On a Mac the setup schedules this for you. On Linux you add one line to cron. Every action and every stop, from every machine, shows up in one place.
What a team gets
- Shared rules. Admins add rules for the whole company or for one team. Every connected machine picks them up on its next sync. Your rules can only add stops; they never turn off the built-in ones.
- Approvals in Slack or by email. When a connected agent needs a person to say yes, a message goes to your team’s Slack channel, and an email goes to the agent’s owner and backup, each with a link. They decide on the dashboard, signed in, so you always know who said yes. A yes works once, for a short time. If no one decides in time, the action stays blocked.
- A second person for company rules. If a company rule stopped the action, the agent’s own owner can’t approve it. Its backup or an admin must.
- A tamper-evident audit log. Rule changes, people changes and every approval are recorded in a chain, so a changed or deleted entry shows. Export every agent action as a CSV.
- Roles. Owners, admins and members. Each agent has an owner and an optional backup approver. Admins manage rules, people and any approval. Members can see everything and approve for agents they own or back up.
- Servers with no one watching. Turn on unattended mode on a server. When the guard stops the same kind of action again, it asks the agent’s owner instead of just failing.
- Log checks for every machine. The Agents page shows whether each machine’s log still passes its tamper check.
What stays on each computer
- The checks run on each computer. The rules decide in milliseconds, and they keep working if the dashboard can’t be reached.
- The full log stays on the computer. The dashboard gets a short line for each action: time, session, tool, a summary of the action, the decision and the reason. The full list is on the Security page.
- Keys, tokens and passwords that Antelog recognizes are removed before anything is written or sent.
- We never upload your code base. The AI check sees only the one action it checks: the command and the text of any script it runs, or the new text of a file write or edit. It doesn’t keep it.
Your security team can read exactly what leaves each computer on the Security page.
Plans
Blocking is free and works the same on every plan.
- Free ($0), one person: the guard, the online dashboard, 10,000 AI checks a month per machine, 14 days of history.
- Pro ($25 a month or $199 a year), one person: 50,000 AI checks a month across your connected machines, 90 days of history, Slack and email alerts.
- Team (custom price): more people, shared company and team rules, approvals with a second person for company rules, the audit log and CSV export. Need single sign-on or your own servers? Ask us.