Skip to content

AgentZ Documentation

Welcome to AgentZ

AgentZ runs AI agents in a sandbox you configure. A sandbox is a reusable set of models, packages, tools and allowed hosts. Each agent uses your own model and can reach only the hosts you list, plus the AgentZ services. An agent holds a placeholder in place of each secret. You set it up once, then chat with the agent.

Start the Quickstart Open AgentZ

Start Here and Reach Your First Chat

You finish with a working agent and your first reply. Have a GitHub or Google account ready, plus an API key from a model provider.

Introduction
What AgentZ is, and the five words you need.

Create a Workspace
Sign in, then create the space your team works in.

Add a Model
Connect the AI model your agents use.

Create a Sandbox
Choose what an agent can use and reach.

Create an Agent
Name an agent and pick the sandbox it uses.

Send Your First Chat
Ask the agent for something and read the reply.

Go Deeper Once Your First Agent Works

Concepts
How organizations, workspaces, sandboxes and agents fit together.

Configure
Models, packages, MCP connections, secrets, skills and custom tools.

Automate
Workflows, schedules, webhooks and dashboards.

Administer
Users, roles, sharing, API keys and the audit trail.

Guides
Worked examples, including a weekly HR application intake.

Reference
Glossary, limits, troubleshooting and the FAQ.

See AgentZ in Action

An AgentZ dashboard with charts for open incidents, revenue by channel, pipeline by region and support load

An agent builds a dashboard when you ask for one in chat.

The workflow graph viewer showing a weekly HR application intake workflow with labeled branches

A workflow is a saved graph. The viewer shows its nodes and branches.

A trace of a custom tool call showing the tool arguments and the tool result

Lens traces show the arguments and the result of each tool call.

Found a mistake or a gap? Open an issue.