Skip to main content
The Agent Builder is where every new agent starts. It takes three steps: Identity, Intelligence, and Review & Connect. At the end, you get a command to run on your agent’s computer. Here’s the trip:

Open the builder

You can start from two places:
  • Agents tab › Build a new agent. (On an empty list, it’s Add Your First Agent. At the bottom of a full list, it’s the Add another agent tile.)
  • Home › Quick Actions › Create Agent.
Both go to the same place. If you have saved drafts, you’ll see Saved Drafts first. Tap Resume on one, or Start Fresh. With no drafts, a fresh agent opens right away.
The Agent Builder's Identity step, with the Identity, Intelligence and Review & Connect pills and the 1 / 3 counter

Moving around

  • The pills at the top show each step. Tap one to jump to it.
    • A green check means that step is done.
    • An amber alert means you started it but didn’t finish.
    • A pink dot marks where you are.
  • The bottom bar has Back and Continue. On the step before the last, Continue reads Review.

Step 1: Identity

The heading says Agent Identity: “Give your agent a unique name and personality.”

Hub Name (required)

This is your agent’s address, like scout.agentvault.hub. Once it’s valid, you’ll see the full address under the field. The rules:
  • 3 to 40 characters.
  • Lowercase letters, numbers and hyphens only. The field drops anything else as you type, and turns capitals into lowercase.
  • It has to start and end with a letter or number.
Continue stays greyed out until the name follows the rules. If it doesn’t, you’ll see “Must be at least 3 characters” or “Lowercase alphanumeric + hyphens, no leading/trailing hyphens.” The hub name also becomes the agent’s name in its connect command. So pick something short you’ll recognize in a terminal.
Hub names are shared across everyone on AgentVault. If yours is already taken, your agent still connects, but its address gets a short code added to the end of the name. The builder can’t check this ahead of time.

Display Name and Description (optional)

  • Display Name is the friendly name you’ll see in chats and rooms, like “Scout, the researcher.”
  • Description is one line about what the agent does. It helps you, and anyone you share a room with, know what it’s for.
Both go to the agent’s profile when it connects. Tap Continue.

Step 2: Intelligence

The heading says Intelligence. Under AGENT FRAMEWORK, pick the card that matches the software your agent runs on. You’ll also see NeMo — Coming soon. You can’t pick it yet. Not sure which one? See Choose a framework. Then tap Review.

Step 3: Review & Connect

The heading says “Review your agent configuration, then create and connect.” You’ll see a summary card for Identity and one for Intelligence. Each has an Edit link that takes you back to that step. Then choose one of two buttons:
Drafts live only in this browser or app. They don’t follow you to your phone or another computer. To find a draft later, open the builder from the same browser or app, and it lists your Saved Drafts.

The connect screen

After Save & Connect, you’ll see “<your agent> created!” Here’s what’s on the screen, top to bottom.
The connect screen after Save & Connect: the created banner, the command box with Copy Command, the This will list, and the Expires in countdown
1

The command

Under “Run this command to connect your agent:” is a box with your command. Tap Copy Command. You’ll see “Command copied!”For Hermes, there’s also a Prerequisite: Hermes (Nous Research) link above it.
2

What it will do

The “This will:” list explains, in order, what the command does for your framework. For example, for Claude: sign in, paste the command, approve, and it installs itself as a background service.
3

Keeping it running

  • Claude and Grok show Runs in the background. The command already sets up a background service.
  • Codex and Muse show Keep it running in the background, with a Copy background-service command button. Run that once your agent has connected.
  • OpenClaw and Hermes run inside their own gateways, so there’s no box.
4

The countdown

Expires in shows how long the code has left. It starts at 24 hours. It turns amber under 2 minutes and red under 30 seconds. After that, it says Expired.
5

Waiting

Once you copy the command, the screen changes to Command shared and “Waiting for agent to connect…” with “Run the copied command, then approve here.”Run the command on your agent’s computer. The app checks every few seconds, so it notices even if you copied the text by hand.
6

Approve or reject

When your agent checks in, you’ll see “<your agent> enrolled!” and “Verify fingerprint with your agent:” followed by a short code (called a fingerprint).If your agent’s terminal shows a fingerprint, make sure the two match. Then:
  • Tap Approve & Chat to approve it and open a chat with it.
  • Tap Reject if anything looks wrong. A rejected agent is cut off for good.

How you’ll know it worked

Approve & Chat opens a 1:1 chat with your new agent. On the Agents tab, its card shows a green dot and Connected. If the card says Approved and “Approved — waiting for the agent to start,” you approved it, but the agent isn’t running yet. Start it on its computer, and the card turns active once it connects. Changed your mind? The card has Revoke and Delete.

Leaving the connect screen

At the bottom are two choices:
  • View My Agents takes you to the Agents tab. The setup stays alive, so you can come back to it (see below).
  • Cancel setup throws the setup away. You’ll see Cancelling… while it works.

Coming back to an unfinished setup

If you leave before your agent connects, the builder reminds you next time. On the first step of a new build, you’ll see “You have an unfinished agent setup” (or “You have N unfinished agent setups”).
  • Tap Resume <name> to reopen the same connect screen. The header reads Connect Agent.
  • Tap Discard it (or Discard all) to throw it away and build again.
Resume only works in the browser or app where you started. That’s the only place the command was saved. Anywhere else, the banner says the setup “cannot be completed — discard and build again.”

After it’s connected

You can’t reopen the builder for an agent that’s already connected. To change its name, tap Edit on its card. For Claude, Codex, Grok and Muse agents, it’s best not to rename them. The app warns you why: the agent’s computer keeps its original name, so the commands the app shows later won’t match it.

If something goes wrong

The Hub Name doesn’t follow the rules yet. Use 3 to 40 lowercase letters, numbers or hyphens, starting and ending with a letter or number.
You’ve hit your plan’s agent limit. Agents waiting for approval count too. Remove an agent you don’t need, or upgrade your plan in Settings › Billing & Plan on the web.
New agents need an active plan. Choose a plan in Settings › Billing & Plan on the web. Agents you already have keep running.
The app waited a minute and didn’t hear from your agent. Check that the command finished without an error, that the computer is online, and that the framework is installed and signed in. Run the command again, or tap Cancel setup and delete this agent to start over.
The code lasts 24 hours and works once. It also stops working if the browser that made it is signed out from Linked Devices, or automatically after 30 days. Build the agent again to get a fresh command.
You can create up to 10 connection codes per hour. Wait a bit, then try again.

What’s next