AI Agent Assurance for your agents

Choose your agents. Keep one company policy⁠.

Scout finds the agents on each endpoint, and company policy answers each step a supported agent takes before it runs.

Agent choice

Let teams switch agents without redoing security⁠.

Write the policy once and each supported agent follows it, so a new agent or a new vendor does not mean a new set of rules.

  1. Google held #1 from to (Gemini 2.5 Pro)
  2. OpenAI held #1 from to (GPT-5)
  3. Google held #1 from to (Gemini 2.5 Pro)
  4. Anthropic held #1 from to (Claude Sonnet 4.5)
  5. Google held #1 from to (Gemini 2.5 Pro, Gemini 3 Pro)
  6. Anthropic held #1 from to (Claude Opus 4.6, Claude Opus 4.7, Claude Fable 5, Claude Opus 5.5)
  7. Google held #1 from to (Gemini 4 Argon)
Best text score per provider at each published leaderboard, axis from 1,380. Source: LMArena leaderboard history (opens in new tab)

Coding agents

Give developers the coding agents they want⁠.

Developers pick the agent that fits the task, and company policy answers each risky step before it runs, whichever agent they pick: pause it, steer it, or block it.

More on coding agents

In the terminal and the IDE

Browser agents

Let agents act in signed-in sessions⁠.

Browser agents work where people are already signed in. Scout shows each provider account in use, sanctioned or not, and company policy answers each step before it leaves the endpoint.

In browser tabs and signed-in sessions

  • Claude for Chrome
  • ChatGPT
  • Chrome
  • Edge
  • Safari
  • Firefox

Desktop agents

Let agents drive desktop apps⁠.

Desktop agents open files and operate applications outside the browser and the terminal. Certiv sits on the endpoint where they act, and company policy answers each step before it runs.

More on endpoint AI agent security

In desktop applications

  • Claude Desktop
  • Codex Desktop
  • Kimi Work
  • ZCode (Z.ai)

Custom agents

Ship your own agents with policy built in⁠.

Home-built agents are a growing category, and company policy applies to them on any framework, with any model, or as plain scripts, the same way it applies to a packaged agent. Certiv sits on the endpoint, so a home-built agent needs no hooks and no vendor integration.

Four home-built agents on one computer: a Python script, a Node service, an MCP server, and a shell script, each on its own puck inside a blue ring. A blue frame runs around the whole machine, and a small Certiv plate sits on its front edge.

How teams build their own agents

  • Python
  • Node.js
  • Shell scripts
  • MCP servers

Claude Code

Say yes to Claude Code⁠.

Run Claude Code at full speed. Company policy checks each step on the endpoint, outside the agent, before it runs.

Claude Code security

A terminal and a code editor standing like screens on a plate, with an orange accent line and a small Claude Code mark.

The agent and its model

Codex

Say yes to Codex⁠.

Let developers run Codex on an OpenAI model or a local one. Company policy checks each step on the endpoint before it runs, and the console lists what policy did.

Codex security

A Codex app window standing behind a command-line window, with a small Codex mark on the plate.

The agent and its models

Get started

Pick the agents. Keep company policy on the work.

Bring the agents your teams already use. Certiv shows you each one and holds supported agents to one company policy.

Book a demo