AgentZ Documentation
Build, run, and govern production agents¶
AgentZ is a zero trust platform for agentic AI. Every agent runs inside a default-deny sandbox, never holds a secret, and writes every step to a replayable trace. Run it on SaaS, on your own cluster, or fully air-gapped.
Self-host in one command
git clone https://github.com/accuknox/agentZ && make deploy
Deploys the manager, gateway, secret-injection sidecar and observer onto your cluster. Prefer the hosted route? Sign up and skip the install entirely. Both paths are covered in Deployment Options.
What is AgentZ?¶
Zero trust AI agents you can deploy anywhere.
Agent platforms usually treat isolation as a setting you switch on later. AgentZ runs every agent inside a default-deny sandbox from the first run. The agent reaches only the destinations you allowed, receives no credential it could leak, and leaves a record of every call it made.
Three ideas carry the whole product:
- Default-deny networking. An agent reaches only what its sandbox permits. Everything else is blocked, and every block is recorded.
- Zero-trust secrets. The agent container never receives a secret. A proxy substitutes the real value at request time, so the model never sees it.
- Control over what agents can run. Define the tools and packages once, reuse that definition across agents, and update it in one place.
It is an AI platform with security built in, and it is not limited to security work. See Beyond Security for what other teams build with it.
Your first workflow, in five steps¶
- Sign up Signing up creates an organization and makes you its super admin. There is no separate org-creation step, and no one has to provision you.
- Create your first workspace Pick a type: General, or AI-SOC. An AI-SOC workspace surfaces integrations, triggers and investigation views a General one does not. See Core Concepts.
- Connect your first integration Choose a provider, authorize once over OAuth, and wait for it to go Ready. Then set each tool to Blocked, Needs Approval, or Always Allow. See Connecting Integrations.
- Create and run your first workflow Add a skill, wire the steps, and run it from chat, the API, the CLI, or a schedule. See Creating Workflows.
- Read the run graph Every model call and tool call the workflow made, in order, down to the token. See Governance and Security.
What to explore next¶
The AgentZ Mental Model
One hierarchy that explains the whole platform. Read this second.
Core Concepts
Agents, workflows, skills, sandboxes, credentials, workspaces and triggers.
Building and Managing Skills
Write one, upload one, or describe the job and let AgentZ write it.
Governance and Security
Zero secret exfiltration, approval gates, and the audit trail.
Deployment Options
SaaS, on-premises, air-gapped, and bringing your own model.