Fort documentation
Understand the boundaries.
Trust what happens inside.
A home for agents that run on their own, inside your company’s cloud, under rules you set.
Fort combines agent identity, policy enforcement and evidence. Every agent gets its own sandbox. Warden checks its tool calls. You decide which actions can proceed and which need a person.
The foundation
Know the agent. Control the agent.
Trust the interaction.
Agent identity & accountability
Each agent has a unique identity, authenticated on every action. Know who owns it, which version is running, and who approved its actions.
Who is this agent, who owns it, and can we trust it?
Governance & protection
Define which tools and data an agent may use. Require a named person to approve risky actions. A manipulated agent is still constrained by its policy; a stop switch can stop one agent or all of them.
Is this agent acting within its authorized boundaries?
Secure interaction & audit
Provider credentials are added outside the agent. Calls to tools are checked against policy, and allowed and blocked actions leave a tamper-evident record.
Can we trust the counterparty and the interaction?
Containment & enforcement
The sandbox keeps the agent in.
Warden decides what gets out.
Fort runs each agent in an isolated sandbox built on NVIDIA OpenShell. Network access is restricted to the destinations you declare. Governed tool calls go through Warden and Fort’s relay before reaching tool servers.
Warden is the part of Fort that decides and enforces what agents may do. It combines policy evaluation, tool-call enforcement, human approvals and the stop switch.
Watch Warden decide