Docs
What ZYBE is
Your developers use AI tools that can run commands and read files on their laptops. ZYBE shows you every one of those tools, tells you what they do, and stops the dangerous things before they happen.
The problem in one minute
Developers now use AI tools that do more than answer questions. Claude Code, Cursor, Codex, Copilot, Gemini CLI and others can open files, run commands, install software and reach the internet on the laptop they run on. They do it with the developer's own access. If one of those tools reads a web page with hidden instructions in it, or is simply asked to do something careless, it can read a password file, delete a folder, or send data somewhere, and nobody sees it happen.
ZYBE makes that visible, and makes it stoppable.
What ZYBE does
ZYBE has two parts.
The sensor is a small program you install on each laptop. It finds every AI tool on the machine. It watches what each tool does. And when a tool is about to do something your rules don't allow, it stops it before it happens and tells the tool why.
The console is the website at console.zybe.ai where you see all of it: every laptop, every AI tool, every risky thing found, every action that was stopped. It's also where you decide the rules.
Words you'll see
| Word | What it means in ZYBE |
|---|---|
| Agent | An AI tool that can act on a computer: run commands, read and write files, install things. Claude Code, Cursor and Copilot are agents. A chatbot that only talks isn't. |
| Endpoint | A laptop or server with the sensor installed. |
| Sensor | The ZYBE program on an endpoint. |
| MCP server | A small add-on an agent can call to do a job: read a database, search the web, talk to GitHub. Developers install them, often from the internet. ZYBE lists and can block them. |
| Skill | A set of written instructions an agent follows for a task. Also installed from the internet, also listed and blockable. |
| Rule | One thing ZYBE checks. "Agent read a credential file" is a rule. Each rule is on or off, and the ones that can stop an action are set to Report, Ask or Block. |
| Report | Let it happen, but record it and show it in the console. |
| Ask | Pause and ask the developer first. They see ZYBE's reason and decide. |
| Block | Stop it. The agent is told why and told not to try another way. |
| Profile | A set of rules that applies to some endpoints, some agents, or both. |
| Group | A set of endpoints, like "Finance laptops" or "All Macs". |
| Finding | Something ZYBE found that needs attention. Issues are standing risks. Threats are things an agent did that look like an attack. |
The four words next to every agent
In the console, every agent on every endpoint has one word next to it. It tells you what ZYBE can do about that agent on that machine.
| Word | What it means | What you can do |
|---|---|---|
| Blocking | ZYBE checks every action before it runs and can stop it. | Everything: Report, Ask or Block. |
| Recording | ZYBE sees what the agent did after the fact, but can't stop it. | See and investigate. |
| Seen | ZYBE knows the agent is installed and nothing more. | Know it's there. |
| Silent | ZYBE is connected to the agent, but the agent has run without reporting. Usually it was open when ZYBE was installed and needs a restart. | Restart the agent. If it stays Silent, see Troubleshooting. |
The word is earned, not assumed. ZYBE only says Blocking when it has actually checked actions from that agent on that machine.
| Not allowed | The company has refused this product, and the laptop has taken away its permission to run. | Nothing. It's handled. |
Some apps give ZYBE no way to check their actions one at a time, so they're Seen until you decide. Type the product's name under Rules, Refused products, and every laptop refuses it. See Refuse an AI app.
Where to go next
- How it works: the drawings. Read this before rolling it out.
- Your first 15 minutes: install on your own laptop and watch it work.
- How do I...: the things people come to ZYBE for, one page each.
- AI tools ZYBE covers: what ZYBE can and can't do on each one.