Creating Your Own AI Agents
Create custom AI agents for recurring tasks using the Agent Library, configure system prompts and tools, then reuse them across your workflows
Build AI agents for recurring tasks, use pre-built templates, or customize them to match your workflow.
Key Benefits and Use Cases
Specialized agents are saved, named, and versioned agents designed for a specific task, role, or recurring workflow.
The main advantage: you set up an agent once and can reuse it over and over again.
You can use a custom agent to:
-
Use it in chat
-
Mention it in comments
-
Run it in a task or project
-
Use it in automations
-
Run it on a schedule
-
Share it with your team
What is an Agent?
An agent is a saved AI configuration for a specific task, role, or recurring workflow.
An agent's configuration can include:
- Name
- Description
- Instructions (System Prompt)
- Model
- Reasoning level
- Tools
- Skills
- Connectors
- Files
- Avatar
An agent can then be run in chat, in comments, in tasks and projects, as well as via schedules or automations.
New agents are private by default and only available to their creator and workspace administrators.
How to Create an Agent
You create and manage agents in the Agent Library.
The Agent Library is your central hub for: Agents, Skills, Connectors
You can create new items there and open and configure existing items.
-
Click the Library button in the upper right corner and select the Agents section.
-
Click the + Agent button
-
In the dropdown menu, you'll find two options:
- New Agent: Create a completely custom agent according to your needs
- Use Template: Use an existing template as a starting point for your new agent
Select Custom Agent to proceed with creation.
-
Choose your creation method
- Create with AI (recommended): Get help from AI while creating your agent. awork guides you through a dialog-based setup process. The AI asks questions, gathers requirements, and helps you build the agent configuration together.
- Create manually: Configure the agent from scratch yourself
-
Give your agent a name and description
-
Define the agent's task and instructions
-
Choose a model, tools, skills, and connectors
-
Set an avatar if needed
-
Test the agent and refine its configuration
Changes to instructions are versioned. Earlier versions can be restored without changing the rest of the agent configuration.
Which changes are possible depends on your access level. Editable entries show their configuration and editing options. Items with read-only access can still be used and viewed, but cannot be changed.
Many changes are saved automatically while you edit. There's no separate Save step in these areas.
Agent Version History
An agent's instructions have a version history.
You can view older versions and restore them if needed.
Using Pre-Built Agents with a Template
You don't always have to create an agent from scratch.
The Agent Library has curated templates created by awork. These already include:
-
System prompt
-
Sensible model and reasoning settings
-
Appropriate tools
You can create your own agent from a template and then customize it. Templates serve as a helpful starting point, not as a fixed, final configuration.
Open the Library, click the + Agent button, and select Use Template to create a normal, fully editable agent in your workspace from it.
Available Agent Types
The templates cover typical agency tasks, including:
-
Review briefings
-
Plan projects
-
Check project and portfolio health
-
Review timesheets
-
Prepare meeting follow-ups
-
Prepare campaign strategies
-
Review personas
-
Check brand safety
The templates are useful even if your workspace initially has little relevant context.
Some templates can even help you create the context you need if it's missing.
Customizing Templates
A template is not a fixed agent. After clicking Use Template, you get an agent that you can edit, test, and then refine further.
For best results, you should adapt a preset to your needs early on:
-
Describe the specific task – Explain exactly what the agent should do
-
Add company knowledge – Include company-specific rules and guidelines
-
Add appropriate connectors – Presets don't include external app connectors by default
-
Provide existing context directly – Point to relevant information instead of letting the agent guess
This is especially important for agents like briefing or brand safety agents, which depend on company-specific guidelines.
You can always start fresh with a new version of the original template if you want to go back to the default configuration.
Using Your Own Agent
You can run your own agent in several ways.
In Chat
-
Create a new thread
-
Select a custom agent
-
Write your message
-
The thread uses this agent's configuration
When an agent is selected in a thread, its configuration is used as the default for that run.
Some settings can still be adjusted for the current thread though. However, these changes do not permanently change the agent.
If the agent is removed from the thread, your normal default settings will be used again.
In Comments
You can mention a custom agent with @ in a comment. As soon as you type the first letter, you'll see available agents suggested. Select the agent you want.
In a Task or Project
On tasks and projects, you can select a custom agent from a dropdown field.
When you select the agent, it runs immediately. It is not permanently added as the person responsible.
So for customer-facing language: Run the agent in a task or project.
In Automations
A custom agent can be used as a step in a project automation.
On a Schedule
An agent can run automatically on a schedule.
When running, the agent uses the permissions of the user who created the schedule. Only the creator can edit it.
A schedule:
-
Can run at most once per hour
-
Can contain additional instructions
-
Can optionally receive a project or task as context
-
Is automatically disabled after three consecutive failed runs
Skills for Custom Agents
Skills are reusable instructions or expertise that an agent can use.
You can:
-
Create a skill in the Agent Library
-
Create a skill directly within a custom agent
-
Import a skill from a compatible skill package
Skills from the Library
A reusable library skill belongs to its creator and can be connected to multiple agents.
This way, a workflow can be maintained centrally and used in multiple agent configurations.
Skills Within an Agent
A skill you create directly in an agent belongs exclusively to that agent.
It becomes active there immediately and doesn't appear as a reusable skill in the general library.
For skills, you can set things like name, description, instructions, supporting files, and required access.
Connectors for Agents
Connectors let an agent use external systems via MCP.
A connector and an authenticated connection are two different things:
-
Connector – describes the external system.
-
Connection – is the specific authenticated connection to that system.
A connector can have multiple connections.
Which connection an agent can use also depends on the responsible user's permissions and available connection. A shared agent doesn't automatically get access to a shared user account.
When an agent is used by multiple users, their respective available connections and permissions apply to each of their uses.
Adding Connectors to an Agent
Each custom agent has its own list of connected external tools.
You can:
-
Select an existing workspace connector
-
Create a new connector
-
Choose one or more available connectors via search
A newly created connector is immediately connected to the agent.
Only the connectors connected to the agent will appear in its configuration afterward.
Adding a connector to an agent doesn't automatically create a new authenticated connection.
Available Connectors
There's a curated catalog with around 30 ready-made connectors, including for example:
-
Slack
-
Notion
-
GitHub
-
HubSpot
-
Salesforce
-
Google Workspace
-
Microsoft 365
Most connectors can be connected with a single click. Some require their own API credentials for the respective application.
If an application isn't in the catalog, you can also add another MCP connector via its URL.
Sharing Connectors and Connections
For shared agents, editors can decide how connections are used.
They can choose whether:
-
Each user uses their own connection
-
A specific connection is provided by the agent
Connections themselves have their own sharing logic.
A connection is private by default. It can then be shared with the entire workspace, specific teams, or individual users.
Only the owner of a connection or a workspace administrator can manage or remove a shared connection.
Sharing Agents and Skills
Agents and reusable skills can be shared selectively.
For agents and skills, you can give access to:
-
The entire workspace
-
Specific teams
-
Individual users
There are two access levels:
-
Read and use – The user can use the agent and use the connected skills
-
Manage – The user can change the agent's configuration and sharing settings
A user with access to an agent doesn't need to be separately enabled for each individual skill used within it to use the agent.
FAQ
What's the difference between the awork agent and a custom agent?
The awork agent is the built-in agent for general tasks and works without setup.
A custom agent is configured, saved, and shared for a specific task or recurring workflow.
Do I have to build a custom agent myself?
No. You can start with a curated preset and then edit it.
Can I change a preset later?
Yes. Using a preset creates a normal, fully editable custom agent.
Can I apply a custom agent to a task or project?
Yes. You can select the agent and run it directly in the task or project.
Can an agent run automatically?
Yes. A custom agent can be run in an automation or on a schedule.
Can custom agents use external apps?
Yes. You can connect appropriate connectors to the agent for that.
What's the difference between a connector and a connection?
The connector defines the external system. The connection is the specific authenticated identity you use to access it.
Can I share a custom agent with my team?
Yes. Agents can be shared with your entire workspace, specific teams, or individual users.
