Skip to main content
Hivenora
Sign inGet Started

Introduction

What is Hivenora?

Hivenora is the authorization layer between your AI agents and production systems. It evaluates proposed actions against intent, impact, and policy — and returns a structured decision before the action is executed.

Private Preview. Hivenora is in active development. The API is implemented and usable. SDK packages are not yet published to npm or PyPI. See the Quickstart for the current integration path.

Why Hivenora exists

AI agents can do consequential things: delete customer records, send emails at scale, move money, export sensitive data. Before Hivenora, there was no structured checkpoint between an agent deciding to do something and it actually happening.

Permissions alone are not enough. An agent that can delete contacts can also delete 48,000 contacts when instructed to delete 12. The difference is blast radius — and blast radius requires understanding context, not just capability.

Hivenora provides that checkpoint. Call evaluate() before any consequential action. Hivenora returns a decision: allow, require approval, or block. Every decision is recorded. Approvals appear in the Control Room for human review.

Where Hivenora sits

Hivenora is SDK-mode infrastructure. It lives inside your application, in the execution path of your agent:

Your Application
│
▼
AI Agent
│
▼
Hivenora SDK → POST /v1/evaluate
│
▼
Decision: allow | require_approval | block
│
▼ (if allowed)
Production System (CRM, Payments, DB…)

The Hivenora SDK calls the Hivenora API (POST /v1/evaluate). Your application receives the decision and acts on it. The agent never directly calls the protected tool — your application code enforces that boundary.

SDK mode trust boundary. In SDK mode, the enforcement boundary is your application code. If your code can bypass the SDK and call the tool directly, SDK mode alone does not prevent that. See SDK vs Enforcement for a precise explanation of what this means for your security model.

Core concepts

Action
A proposed operation the agent wants to perform. Described by a string identifier like crm.delete_contacts.
Intent
The natural-language goal the user gave the agent. Used to detect when the action scope exceeds what was requested.
Context
Structured metadata about the action: records affected, financial amounts, environment, reversibility, data sensitivity.
Blast Radius
An assessment of potential impact: critical / high / medium / low. Computed deterministically from context.
Policy
A rule that maps actions and context to decisions. Defined in the Control Room per agent.
Decision
The output of evaluation: allow, require_approval, or block. Always paired with reasons.
Shadow Mode
Hivenora evaluates actions but does not enforce decisions. Used to validate policies before enabling enforcement.

Next steps

Quickstart →