Transcript
Your documents are useful to people. They are just as useful to AI agents — if you can control what those agents are allowed to touch. That control lives in Agent Manager.
Opening Agent Manager
Agent Manager is the agentic layer of DocReveal. You open it from the menu on the right.
What Agent Manager does
From here you connect any AI agent — Claude, your own, or a partner product — to your workspace. Four tabs cover the whole lifecycle: Connect gives you the endpoint, Capabilities decides what agents may do, Activity is the audit trail, and Keys holds the credentials.
Three steps to connect
Connecting an agent takes three steps. First, create an access key — you will see the secret only once. Second, paste the DocReveal address into your agent. Third, ask the agent a question about your documents. Setup guides cover Claude Code, Claude Desktop and custom agents; ChatGPT support over OAuth is coming.
Capabilities and guardrails
Before you create a key, look at Capabilities. Each capability groups the actions an agent can call, and you can open it to read every one of them. Capabilities marked sensitive carry a second lock: the agent must confirm the action against a preview, or a person has to approve it. Team members is the strictest of them — managing your users always waits for a human.
Creating the key
Now the first step. In the Keys tab, choose Create key. Give the key the name of the agent it belongs to — that name follows every call this agent makes into the audit trail, so a vague name costs you later.
Only what the agent needs
Grant only what the agent needs for its job. This agent reads documents and answers questions, so mapping, data models, transcription and team members come off the list. Set an expiry, and create the key. Anything you leave unchecked is not merely refused — the agent never sees that it exists.
The secret is shown once
Copy the secret now. DocReveal keeps only a hash of it, so this is the one moment it is readable. From here the key appears in the table with its prefix, its capabilities, and when it was last used.
Rotate and revoke
Treat the key as you would any production credential. Rotate it on a schedule — the old secret stops working the moment the new one is issued. And when an agent is no longer in use, revoke it.
One key, one agent, one clearly drawn boundary. Next, put that agent to work.