Policy control for AI agents

Secure yourAI agents

Night Agent adds a policy layer between AI agents and the computer, so teams can control execution before work reaches the host.

What you control

  • Allow, block, ask, or sandbox
  • Runs locally on the computer
  • Works with Claude Code, Codex, and terminal workflows

A simpler way to control agent activity

Give teams visibility, control, and auditability without forcing them through implementation detail first.

See what agents are trying to do

Night Agent intercepts commands and tool calls before execution, so teams have visibility before anything touches the computer.

Apply policy before execution

Use readable YAML rules to decide what is allowed, blocked, confirmed manually, or isolated in Docker.

Keep a clean audit trail

Every decision is recorded with context, risk signals, and integrity checks so review does not depend on trust alone.

How it works

The core workflow stays simple: intercept, decide, record.

1. An agent tries to act

Claude Code, Codex, or a shell workflow tries to run a command or tool call.

2. Night Agent evaluates policy

It checks the request against deterministic rules before the command reaches execution.

3. The right outcome happens

Safe work continues, risky work is blocked or confirmed, and isolated work moves to sandbox.

Start local. Add visibility when you need it.

Start with local enforcement through the CLI, then add a browser layer for shared visibility and policy management.

Night Agent CLI

Local control layer

The open source CLI runs on the computer, intercepts agent activity, applies policy, and writes a signed audit log.

See product details

Night Agent Cloud

Browser visibility for teams

Cloud adds project-level visibility on top of the local CLI without moving execution away from the computer.

  • Live event dashboard
  • Searchable event history
  • Policy editing from the browser
  • Multi-project oversight
See Cloud

Built for the workflows teams already run

Claude CodeCodexTerminal agentsYAML policyDocker sandboxSigned audit log

Start with local control. Expand to team visibility.

Use the CLI to govern execution on the computer, then add Cloud when teams need shared oversight across projects.