Omnet

AI agent security

Let AI agents work.Approve what matters.

One wrong agent command can delete a repo, leak a credential, or push code no one reviewed. AgentWall pauses risky supported actions from Claude Code and Codex until you approve them — locally, in plain English.

Download the free betaSee what it covers

Ubuntu 24.04 (amd64) · macOS Apple Silicon · Windows 11 x64 · Public beta

AgentWall — Decision log

Live

  • rm -rf ~/projects/data

    Blocked

  • git push --force origin main

    waiting for your review…

    Ask

  • npm run test

    Allowed

  • curl api.internal/tokens

    Blocked

decisions stay on your machine

Why AgentWall

Agents don’t pause to ask.AgentWall does.

AI coding agents edit files, run commands, and call tools on your real machine — at machine speed. A single bad action can destroy work, expose secrets, or ship something no one reviewed.

AgentWall adds the missing moment of control: a checkpoint where risky supported actions wait for a human decision — with the reasons in plain English.

How it works

Three steps to a safer agent.

01Installed

Connect

Install AgentWall and connect a supported agent — Claude Code or Codex — through its hook interface.

02Ask

Checkpoint

Risky supported actions pause at a clear decision point before they run.

03Allowed

Decide

Allow or block with context, and review the decision history — locally, in plain English.

The product

BetaDeveloper Security

AgentWall

Give your AI agent a clear safety check before it changes files, runs commands, or reaches selected tools.

  • Review risky actions
  • See decisions in plain English
  • Stop a supported agent before it acts
  • Know where coverage stops
Get AgentWall freeAll products

  • Ask before risky actions

    AgentWall can pause a supported agent before it runs a risky command, changes files, or uses a selected tool.

  • Keep decisions visible

    See what AgentWall checked and what it allowed, so your team can understand what an agent was trying to do.

  • Make safer calls quickly

    When AgentWall needs you, it explains the request clearly so you can approve it or stop it.

  • Know when it needs attention

    AgentWall tells you when its supported connections are not working properly, instead of pretending you are protected.

Trust, verified

Security claims you can check.

We publish what AgentWall covers, what it doesn’t, and how to verify every release.

Capability matrix

Every supported integration, its maturity label, and the evidence behind it.

View the matrix

Signed releases

Published downloads with verification files and a minisign public key you can check.

View downloads

Documented bypasses

The ways AgentWall can be worked around, written down instead of left out.

Read the limits

security.txt

The standard address for reporting a vulnerability to Omnet.

Open security.txt

Enterprise

Give your team a safer way to use AI agents.

AgentWall Enterprise is planned for teams that need simple, shared rules and a clearer view of how supported AI agents are used. It is not available yet.

  • Central policies

    Set simple rules for the supported AI agents your team uses.

  • Fleet visibility

    See which devices and supported agents are connected and need attention.

  • Clear activity history

    Review important approvals and blocked actions when your team needs answers.

Talk to us

We reply by email.

AgentWall beta

Put a checkpoint in front of your agents.

Download the free beta for Ubuntu, macOS, or Windows — or talk to us about your team.

Download the free betaTalk to us