How Reagent Works
The Reagent lifecycle at a glance, and the terms used throughout these docs.
Reagent gives you one place to govern every AI tool your teams adopt and use, from procurement to production. It connects to the tools you already run, adds no sensors of its own, and turns what those tools can see into an AI inventory you can assess, enforce policy on, watch, and pay for with your eyes open.
The lifecycle
Reagent runs AI Adoption Security as one loop. Each stage is a page in the Reagent sidebar.
| Stage | What it does |
|---|---|
| Connect | Connect your identity provider, EDR, AI tools, gateways, cloud accounts, network, and developer platform. Every tile is self-serve. |
| Discover | Inventory every AI in the building, sanctioned or shadow, and attribute each one to a person and team. |
| Assess | Understand the risk and value of every AI product, and which controls hold that risk down. |
| Enforce | Publish your AI acceptable-use policy and enforce it where agents actually act: on the device, at runtime, and at the gateway. |
| Monitor | Watch agent behavior, escalate what matters, and respond with the authority you grant. |
| Optimize | See AI spend by tool, team, and person, set budgets, and reclaim what is not used. |
These docs cover the first stage in depth. What you connect determines what every other stage can see, so the setup guides under Connect Your Stack are where most of the detail lives. The other stages have their own guides inside the product.
Terms used in these docs
- Discovery signal. A category of AI usage Reagent can inventory, such as AI app sign-ups and OAuth grants, AI desktop apps, local LLM runtimes, or MCP servers. Each connection lights up one or more signals, and each setup guide says which.
- Live-validated and template. A live-validated signal has been confirmed on a real environment and starts watching as soon as the connection is active. A template signal ships with the connector and stays paused until you confirm the data source on your environment.
- Alias tile. A tile on the Connections page that reuses another connection's credential. Connect the primary tile, grant the one extra permission the alias needs, and it activates on its own.
- Lake. The environment's data lake, where connected telemetry lands and can be searched. Connectors that push telemetry from developer machines require it to be active.
- ABBA. Agent and Bot Behavioral Analytics: Reagent's behavioral monitoring for AI agents.
- Response action. A containment or coaching step Reagent can take through a connected tool, using the write permissions you chose to grant when you connected it.
- Forward Deployed Engineer (FDE). A named Zaun engineer who works alongside your team on rollout and policy. Connection setup does not depend on one.
Security posture
- Every environment is isolated, with its own data lake, credential store, and configuration.
- Credentials and data are encrypted at rest and in transit. Credentials are scoped to the environment they were saved in.
- Role-based access control with granular permissions, custom roles, and SAML group mapping from your identity provider.
- Personal data in findings is masked by default. Revealing it is a confirmed, audited action.
- Every platform action is audit logged. Data retention is configurable per environment.
- SOC 2 Type II.
Where Reagent runs, including inside your own boundary, is covered on Deployment.