Start free trialFree trial
Security

Your machines. Your rules. Every action logged.

agnte runs agents on hardware you control, in sandboxes you configure once, with a journal of everything they do.

How does agnte keep AI coding agents safe?

agnte runs every agent in a sandbox on your own machines, applies one central command policy to every machine, never gives agents the GitHub token, stores secrets in the OS vault and journals every call and command with its actor. agnte provides audit-ready logs; it does not hold a compliance certification today.

Controls

Security controls, live today

01

Sandbox on every machine

Linux agents run as a non-sudo user in bubblewrap with LAN blocked by nftables. Windows agents run in locked-down Docker with an egress proxy.

02

Self-tested

The sandbox self-test confirms sudo fails, /etc writes fail, SSH keys are unreadable and LAN connections are refused.

03

Central command policy

73 denied commands, denied paths and research-only web access, enforced by a pre-tool hook and logged.

04

Agents never hold the GitHub token

Only the control layer writes to GitHub.

05

Scoped access

Owner, operator and read tokens across all 110 API routes; destructive actions need confirm or dry run.

06

Encrypted secrets

Keys live in the OS vault (DPAPI) and are shown as hints only.

07

Audit-ready journal

Every agent call and command journaled with its actor: 1.42M events in production.

08

Backups

A database backup with an integrity check on every upgrade.

FAQ

Security questions

Is agnte SOC 2 or ISO certified?

No. agnte does not hold a compliance certification today. It keeps audit-ready logs: a journal of every agent call, command and decision with its actor.

Where does my code run?

On your own machines: your PC and the Linux VMs you add over SSH. Each agent runs in a sandbox on that machine.

Can an agent leak my credentials?

Agents cannot read SSH keys, do not hold the GitHub token, and secrets are stored in the OS vault and shown only as hints.

Can I restrict what agents run?

Yes. A central command policy with 73 denied commands, denied paths and research-only web access applies to every machine from one screen.

Keep agents inside the lines you set.