Least-privilege scoping.
An agent gets exactly the tools its job needs — and not one more.
// policy
role: support-agent → scope [mail.send, crm.read]
A raw key is all-or-nothing, and a confused or hijacked agent's blast radius is exactly the set of things it can reach. Least privilege keeps that set as small as the job allows.
Grant an agent the precise set of actions its job calls for, packaged as a reusable role. Anything you don't grant never shows up in its tool manifest — it can't call or even discover it.
Intercept
The agent attempts an action. HiveKey catches it in the path — nothing reaches the tool yet.
Evaluate
Scope is defined once as a role and applied across the fleet — read-only by default, with every write a deliberate grant rather than the starting point.
Enforce & log
The verdict is enforced — allow, block, or route for approval — and written to the audit trail, attributable to the agent's owner.
Agent
attempts an action
HiveKey
scope · guard · log
Tool / MCP
only allowed actions
Built for security and platform teams.
The smallest possible blast radius per agent
One role, defined once and applied across the fleet
Ungranted tools are invisible to the agent, not just blocked
Least-privilege scoping is one expression of Scope.
Every capability rides the same spine — Scope what an agent can do, Guard each action in the path, Log all of it on one trail.
Enforce every action your agents take.
Scope, guard, and log every action — and enforce it in the path, before anything happens.