Get Started
This page takes one agent through HYPR AgentPass™ end to end, so you can see what the gateway does with its traffic. Each section links to the guide that covers the task in full.
Keep the first run small: one connector, one agent and one supervisor. By the end you have a request you can find in the log and an approval you answered on your phone. Add breadth afterwards.
Start Here by Role
Pick the starting point that matches your job.
| You are | Start with |
|---|---|
| An administrator implementing a use case | A playbook: See and govern the AI tools already on your workstations, Give a team governed coding agents, or Require a person for high-value agent actions |
| An administrator setting up the gateway for the first time | The sections on this page, in order |
| A supervisor of one or more agents | Being a Supervisor |
Check What You Need in Place
Settle the gateway's public origin first. Connectors and identity providers are configured against it.
Confirm you have:
- The gateway's public origin, reachable from outside your network
- An OpenID Connect provider you can register an application with
- Whatever your chosen connector needs on its own side
Sign In and Connect Your Identity Provider
Console access comes from your own identity provider, and you configure it in the console itself. Do this before you invite anyone else, so they arrive to a working sign-in.
You will:
- Register the console's redirect URI with your provider
- Enter the issuer URL and client credentials
- Decide whether roles arrive from a provider claim or are assigned by hand
Go to Connect the Console to Your Identity Provider →
Give the People Involved Roles
A role decides what someone can do in the console and which tools their agents may use. You need at least one supervisor before you can onboard an agent, because onboarding waits for a supervisor's approval.
You will:
- Review what the out-of-the-box roles grant
- Assign a role to at least one supervisor
Go to Assign a Role to Someone →
Add One Connector
A connector is a target you bring under the gateway. Pick the one your first agent needs, and leave the rest for later.
You will decide:
- Which kind matches your target: an AI provider, an MCP server, a REST API, a cloud agent or a channel
- Which provider-side setup that kind requires first
These connectors have full walkthroughs:
- AI providers: Anthropic, OpenAI, AWS Bedrock and AWS Bedrock Mantle
- Microsoft 365 Copilot
- Provider-managed cloud agents: Anthropic and AWS Bedrock AgentCore
Set Up a Supervisor's Approval Passkey
Supervisors can answer approvals on a phone, with a passkey. When your organization requires signed approvals, the passkey must exist before an agent can be approved. If you are also the supervisor, set up your own first.
You will:
- Connect the gateway to your HYPR tenant, once, as the administrator
- Install the HYPR app and sign in
- Enroll a passkey from the gateway on your phone
Go to Connecting to Your HYPR Tenant →, then Setting Up Your Approval Passkey →
Onboard Your First Agent
Point one agent at the gateway. Its first request is held for approval instead of passing straight through. That hold is what you are here to see. The hold needs the connector's onboarding mode set to Require Approval, the default for Anthropic and OpenAI connectors. An AWS Bedrock connector starts in Observe only.
You will:
- Point the agent at the gateway URL
- Approve its first connection
Find the Request in the Log
The Request Log shows the traffic your agent just sent. Each entry is attributed to the agent and shows the model it used and how the request ended.
You will:
- Filter the live stream to your one agent
- Inspect a single request
Go to Viewing the Request Log →
Make One Action Need a Person
Onboarding holds an agent once. A policy holds it again for the specific actions you care about.
You will:
- Create one policy and associate it with your connector
- Have the agent attempt the action, then answer the approval on your phone
Go to Writing a Policy by Hand →
You can also describe the rule in plain English and let the gateway draft it. See Authoring a Policy with AI →. That path needs a drafting model configured first, in Setting Up AI Authoring.
To see what arrives on the phone and what each answer does to the agent, go to Approving and Denying Agent Actions →.
Where to Go Next
With one agent working end to end, choose your next direction.
| If you want to | Go to |
|---|---|
| Implement a complete use case, step by step | AgentPass Playbooks |
| Govern what the model may be asked or told | Governing Inference Traffic |
| Limit which tools a role's agents may call | Controlling MCP Tool Access by Role |
| Understand how agents, supervisors and delegation are identified | Identity |
| Give a non-human caller its own identity | Service Agents |
| See the AI tools already in use on your fleet | Discovering Unmanaged Agents |
| Keep workstation agents on the gateway | Enforcement Points |
| Control what the gateway retains | Data Management |
Related
- HYPR AgentPass: Overview — what AgentPass does and who uses it
- HYPR AgentPass: How It Works — the request path these sections assemble
- Prerequisites — the full list behind the first section