Build Your First Co-Worker

This guide takes you from an empty workspace to a live Digital Co-Worker you can chat with — a simple support assistant on NeoGPT. It takes about ten minutes.

For everything each tab can do, see the Digital Co-Workers section. For a richer, multi-channel build with skills, memory, and human approvals, see the worked example.

Before you start

You need access to a workspace with permission to create and publish Co-Workers, and at least one foundation model enabled in the workspace. If either is missing, ask your workspace administrator.

Step 1 — Create the Co-Worker

In the left sidebar, open Coworkers and click Create Coworker. On the welcome screen, click Let’s get started.

Step 2 — Describe who it is

On Basics & Identity, fill in the five required fields:

  • Name — Ada

  • Tone — Friendly

  • Job role — Support Assistant

  • Foundation model — any model enabled in your workspace

  • Objective — Answer common questions clearly, and hand off to a human when unsure.

Optionally add a Persona & principles note such as “Be concise and always offer a next step.” Then Save — your Co-Worker now exists as a draft.

Step 3 — Give it a channel

Open the Channels tab and add NeoGPT. Give it a name and a welcome message such as Hi! I'm Ada. What can I help you with?. NeoGPT is internal chat and needs no credentials, so it is the quickest way to test.

Step 4 — Add a skill (optional)

On the Skills tab, switch on one ready-made skill your workspace offers — for example a web-search or knowledge tool. If none are available yet, skip this; you can add skills any time.

Step 5 — Publish

Click Publish and confirm. Ada is saved as a version, deployed, and becomes Active — ready to use on NeoGPT.

Step 6 — Try it out

Open Ada in NeoGPT and send a message. Then open the Traces tab on Ada’s page to see how it handled the request — which skills it used and what came back.

What’s next

That’s a working Co-Worker. From here you can:

  • Add more channels — Microsoft Teams or Outlook email.

  • Give it more skills, including agents you have built in other builders.

  • Turn on memory so it remembers across conversations.

  • Add guardrails and approvals for sensitive actions.

For a complete, real-world build, follow the worked example.