Agents let you put Superhuman’s AI to work on the tasks you do most. Instead of asking for help each time, you build an agent once — either through a simple chat or in the agent builder — and then let it act automatically when the right moment arrives across your apps and tabs.
This article walks you through the agent builder from start to finish, so you can create an agent that’s ready to help across your email, docs, and connected apps.
Note: The agent builder is a beta feature, so you may notice it change over time. It’s currently available to Pro, Business, and Enterprise users.
Before you begin
Every agent is built from a few key parts: a name, a description, instructions, triggers, and the tools it can access. You’ll define each one on the agent setup page, and you can return to edit them at any time.
You don’t need to fill in everything at once. Start with a name and instructions, then add triggers and tools as your agent takes shape.
Two ways to build an agent
You can create and update agents in whichever way feels most natural:
- Use the agent builder: Fill in each field yourself for full control over your agent’s name, description, instructions, triggers, and tools. This is the method covered in this article.
- Chat with Go: Describe what you want in plain language, and Go will set the agent for you. It’s the fastest way to get started, and you can open the agent builder afterward to fine-tune the details.
Both methods create the same kind of agent, and you can switch between them anytime.
Access the agent builder
To create your first agent, follow these steps:
-
Type “create an agent” while chatting with Go, or go to go.superhuman.com and click the menu icon in the upper-left corner:
-
Hover over the Agents section, then click the plus icon that appears:
- Click Create an agent.
- The agent setup page opens, where you can define everything your agent needs to run.
Next, let’s go through each section, from top to bottom.
Create an agent
The agent setup page is organized from top to bottom. Work through each section below to define how your agent looks, when it runs, and what it can do.
Name your agent
At the top of the setup page, give your agent a name. Choose something short and recognizable so you can find it quickly in your agent directory, for example, Inbox Triage or Meeting Prep.
Select the icon next to the name to give your agent a visual identity that stands out in the directory.
Add a description
In the Description field, write a short summary of when to use this agent.
Your description does two jobs: it appears in the agent directory so you and your teammates know what the agent is for, and it helps Superhuman route the right requests to the right agent. Keep it clear and specific — one or two sentences that answer the question “When should this agent step in?”
Example: “Use this agent to summarize long email threads and draft a reply in my voice.”
Write instructions
The Instructions field is where you tell your agent how to behave. This is the most important part of your setup, and the more specific you are, the better your results.
Strong instructions usually cover:
- Role: What the agent is responsible for.
- Tone: How it should sound, for example, warm and concise.
- Rules: Anything it should always or never do.
- Examples: A sample of the output you’re looking for.
Write in plain, direct language, just as you’d brief a capable teammate.
Add triggers
Triggers tell your agent when to run. An agent can have more than one trigger, and you can turn each one on or off at any time.
To add a trigger:
- In the Triggers section, click + Add trigger.
- Choose a trigger type from the list on the left.
- Configure its settings, then select Save.
You can choose from the following trigger types:
- Scheduled: Run the agent automatically at a time or interval you set, such as every weekday morning.
- Proactive assistance › While writing: Offer suggestions as you write. These appear as underlines and inline suggestions, similar to Grammarly’s writing suggestions. You can set where the agent helps (Email, Docs, coding environment, and more), choose an underline color, and set the inline configuration to automatic or manual.
- Mail › Email received / Email sent: Run the agent when an email arrives, or right after you send one.
- Slack › Agent mentioned / Message received: Run the agent when it’s mentioned in Slack, or when a message comes into a channel it’s watching.
Note: Proactive assistance triggers appear as underlines and suggestions directly in your writing, so you see the agent’s help right where you’re working.
Connect tools and manage access
In the Tools and Access section, choose what your agent can see and do.
Start with the built-in tools:
- Web access: Let the agent search and read from the web.
- Superhuman Docs: Let the agent read and work with your docs.
- Superhuman Mail & Calendar: Let the agent read and act on your email and calendar.
Turn on each option to grant access.
To connect other apps:
- Click + Add connections.
- Browse the available connectors, such as Google Calendar, Gmail, Outlook Mail, Google Drive, Linear, and more.
- Click Add next to each tool you want to add. Tools marked Connected are already linked to your account.
Beyond the built-in tools, you can connect from a library of around 60 connectors, with more added over time.
Tip: Grant only the access your agent actually needs. This keeps your agent focused and helps keep your data secure. You can always add more tools as its responsibilities grow.
Save and test your agent
When you’re happy with your setup, select Save in the upper-right corner.
Your agent is now live and will run whenever its triggers fire. We recommend testing it once — trigger it manually or wait for its first run — to confirm it behaves as you expect. You can return to the agent builder at any time to fine-tune its instructions, triggers, or tools.
Find and add agents built by your team
Agents that your teammates publish appear in your company’s agent directory, so you don’t have to build everything from scratch.
To find and add an agent:
- Open the agent directory.
- Browse or search for an agent. Each agent’s description tells you what it’s for.
- Select the agent, then add it to start using it.
When you add a published agent, you get your own version to work with. Depending on how it was shared, your version either stays in sync with the original (an install) or becomes your own copy that you can edit freely.
Note: Agents are published within your organization only. There’s no global, public directory, so your team’s agents stay inside your company.
Share and publish your agent
Once your agent works the way you want, you can let others use it. There are two ways to do this, and each one gives people something different.
Share with specific people
Give named teammates access to the agent you built. They work with your agent, not a copy, so everyone stays on the same version. You choose how much each person can do:
- View: See the agent’s setup.
- Use: Run the agent.
- Edit: Change the agent’s setup.
- Manage: Control the agent’s access and settings.
Sharing works best when a handful of people need to work with the same agent.
Publish to your company directory
List your agent in the company directory so anyone in your organization can find it and take their own version. Instead of accessing your agent, each person gets one of the following:
- An install: Their version stays in sync with your original as you update it.
- A copy: Their own version to edit and adapt however they like.
Publishing works best when you want to share an agent broadly across your organization.
Note: Access levels (view, use, edit, manage) apply only when you share with specific people. The install-or-copy choice applies only when you publish. Publishing is limited to your organization, with no global publishing.
Tips for building effective agents
- Start narrow. Give your agent one clear job before you expand its scope.
- Be specific in your instructions. Concrete rules and examples lead to better output.
- Match triggers to the work. Use scheduled triggers for routine tasks and proactive assistance for in-the-moment help.
- Grant the minimum access needed. Add tools as your agent’s responsibilities grow.
- Revisit and refine. Check your agent’s output and adjust as you learn what works.