---
title: "1Password"
description: "Bridge 1Password credentials to your AI agents"
url: "https://gazebohq.com/vaults/1password"
---

1Password can remain the system of record for API keys, OAuth client secrets, and service-account credentials while Gazebo acts as the agent-facing policy layer. This is useful when an agent should use one item but must not browse a shared engineering vault. Start with a dedicated 1Password service account and vault access limited to the items Gazebo must read; avoid using a human account token. Then map individual secrets to agent profiles so access is granted for a task rather than by broad vault membership.

## Why bridge 1Password to AI agents through Gazebo?

A 1Password service account’s vault permissions remain the upstream capability boundary. If it can read every item in a vault, any compromise of that integration path has that blast radius, even though Gazebo can limit normal agent requests to selected mapped secrets. Use a dedicated vault or narrowly scoped service account where practical, monitor both systems, and distinguish access from use: Gazebo records the requesting agent and decision, while 1Password records vault-side activity. Removing a Gazebo profile blocks future brokering; rotate the underlying item and service-account token if either may have escaped.

## How it works

1. Create a non-human 1Password service account with access only to the required vault or dedicated integration vault, and store its token securely.
2. Map exact items or fields to Gazebo profiles; avoid a broad profile that represents an entire vault simply because several agents need unrelated secrets.
3. An agent requests a mapped credential through MCP, and Gazebo checks its profile before retrieving the item from 1Password.
4. Review Gazebo decisions alongside 1Password activity records and the downstream provider’s logs. A retrieval does not establish that the downstream API was called.
5. Disable the agent profile to stop future requests. If a secret value or the 1Password service-account token is exposed, rotate it upstream and revalidate the remaining mappings.

## Use cases

**Single source of truth for credentials:** Keep ownership, rotation, and item metadata in 1Password, while agents receive only the mapped runtime secret through Gazebo. The Gazebo integration must still have upstream vault access, so use a dedicated service account and avoid treating agent-level policy as a replacement for careful vault design.

**Audit AI access to 1Password secrets:** Use Gazebo’s identity and decision record to answer which agent asked for an item, then use 1Password events to confirm the vault-side retrieval and downstream service logs to investigate use. During offboarding, remove the agent profile and rotate any item that could have been retained by its execution environment.