Introduction

Relynt is a multi-tenant authorization, approvals, and signed-receipts gateway for AI agent write actions. It sits between your agent and its target systems, ensuring every write action is evaluated against policy, optionally approved by a human, and recorded with a tamper-evident receipt.

How it works


Agent → Relynt Gateway → Policy Engine → Approval? → Connector → Receipt

Your agent calls the Relynt gateway before performing any write action
The policy engine evaluates the request against your configured rules
If the policy requires it, an approval request is sent to Slack for human review
If allowed (or approved), the connector executes the action against the target system
A signed receipt is recorded for every attempt — whether allowed, denied, or pending

Core concepts

Concept	Description
Organization	Tenant boundary — all agents, policies, and receipts belong to one
Agent	Non-human actor registered under your organization
Policy	Versioned rules that evaluate each request (allow, deny, or require approval)
Connector	Configured HTTP endpoint that executes the actual write action
Receipt	Append-only, signed audit record of every gateway call
Approval	Human decision via Slack when policy requires one

Architecture

Relynt has two planes:

Control plane (Dashboard)

Manage organizations, memberships, and agents
Rotate agent credentials
Create and version policies
Configure connectors
View approvals and receipts

Runtime plane (Gateway)

Authenticate agent credentials
Evaluate policies
Route to Slack for approvals when required
Execute connector calls
Record signed receipts

Next steps

Quickstart — integrate your first agent in minutes
Agents — understand agent registration and credentials
Policies — configure allow, deny, and approval rules
API Reference — full gateway endpoint documentation