AI agents are built into monday.com and act within the context of your work. They monitor activity, make decisions based on your defined rules and priorities, and execute tasks end to end on your boards and workflows.
This article walks you through how to build an AI agent from start to finish: creating your first agent, configuring its instructions and permissions, adding tools and skills, setting up automated jobs, and connecting external channels.
For a full overview of AI agents, see the AI Agents on monday.com article.
Create your first agent
First, click on the AI Agents tab in the left navigation pane. Next, click on the + New agents tab on the left pane menu.
You can describe your ideal agent in plain text, choose a pre-defined starting prompt, or click on + Start from blank at the top right of the page.
You can also use pre-built agents in Agents made for you. These templates are personalized and specifically generated for your role within your company and based on the data on your boards.
Next, you will see the following page. Here you can customize your agent as you wish. To continue, click on Onboard:
Pause and resume your agent
To pause your agent, click the Pause agent button, located in the top right three-button menu of the Agent builder. In a Paused state, you can continue building and editing, but the agent cannot run.
To resume, take your agent live again by clicking the Go live button located in the This agent is paused message at the top of the Agent builder.
Customize with the agent chat
There are several components in the Agent builder that give you the ability to customize, tweak, and manage your agent. All changes you make, whether through the chat or the config panel, are saved and applied automatically.
Use the agent chat on the left side panel to ask the agent builder about your agent's abilities, ask troubleshooting questions, make changes, and upload files or boards as context until you are satisfied with your final result.
The agent chat will guide you through the entire building experience, suggesting improvements, additional context to add, and when to run a simulation to test out the agent. Not sure what can be done? Ask the agent what it can and cannot do directly.
The chat will point out what is missing, what you can add to your agents, and will let you know about any missing information. If you have any questions, you can of course ask, but you do not need to decide on your own how to best optimize your agent. If you are unsure, the chat is there to help.
After the agent performs any configuration action (such as adding an instruction, trigger, or tool), an Undo option appears immediately. The agent also supports full editing and removal of its own configuration.
The Agent builder tabs
The Agent builder is organized into four tabs, each covering a different part of how your agent operates:
- Brain tab: Configure your agent's core setup, including instructions, knowledge and access, tools, skills, and AI model. This is where you fine-tune everything the agent needs to understand its work and act on it.
- Jobs tab: Break your agent's work into separate, focused tasks. Each has its own trigger and its own scoped instructions.
- Channels tab: Define how your agent communicates with people in external apps such as WhatsApp and Telegram.
- Activity tab: Monitor every run your agent has completed, including status, apps used, and AI credits consumed. Covered in the Run and manage your AI agents article.
The sections below walk through the Brain tab in detail, followed by the Jobs and Channels tabs.
Brain tab - Instructions
You can review the instructions the Agent chat has generated for your agent. Make any additions or edits needed directly in the box, or click on Expand to open up the full document to make edits.
Remember that the agent cannot perform actions outside of what it is explicitly permitted to do. Proper configuration will ensure that your agent runs safely and in a predictable manner.
Brain tab - Knowledge and access
Use the Knowledge and access options to grant access to monday boards and other files, increasing context awareness and improving results.
You can also specify the exact permissions you want to grant to the agent, for example whether the agent can only read the board data or if the agent can also edit and add to the board data. You have full control over the permissions granted to the agent, allowing you to be as private or open with the agent as you prefer.
If there is additional context (for example, FAQs, governance and legal information) that's located outside of your monday boards and is relevant for your agent to use, you can upload files within this section. Supported file types are: pdf, doc, txt, excel, and csv.
The agent proactively recommends relevant boards to connect based on your instructions. Recommendations appear in the Knowledge and access section and via the agent chat, with up to 5 suggestions at a time.
Brain tab - Tools
Use the Tools section to maximize your agent's abilities with web search capabilities and connections to platform integrations.
Agents can also connect to external tools through integrations, allowing them to pull in information, update records, and keep systems aligned without manual copy and paste.
To add a tool, describe what you need in the agent chat. The agent will:
- Identify the right tools based on your request
- Ask the questions needed to resolve any ambiguity
- Surface the connection step directly in chat
- Confirm exactly what was configured so you can review before moving on
Tools are classified as Read (search, pull data) or Write (send, create, edit, delete). Read tools are active immediately after connection. Write tools are added as inactive and the agent will prompt you to review the setup. You can explicitly activate the tool when you're ready.
Guardrails define the boundaries of what a tool is and isn't allowed to do. For example, you can restrict the agent to only post in a specific Slack channel, or only email certain contacts. The agent asks these questions as part of the setup flow and applies your answers across all relevant tools at once, so you don't need to repeat yourself for each one.
To use dynamic values in your connections, make sure you define the parameters explicitly in the agent instructions in the Brain tab, or via chat.
Supported integrations include email providers such as Gmail and Outlook, messaging platforms, calendars, CRMs, developer tools, file management tools such as Google Drive, and more. The full list of available tools is always up to date in the Brain tab under the Tools dropdown.
Here you can also:
- Enable web search - toggle access on and off using the switcher to allow your agent access to searching the web for more accurate search results
- Grant access to the AI Notetaker - the agent will have access to the meeting records, summaries, transcripts, and action items.
-
Add a custom MCP - In addition to the built-in integrations, you can extend your agent's capabilities by connecting any external service that exposes an MCP-compatible endpoint. This allows you to go beyond the built-in toolset and connect services you control directly to your agent. Under tools, click on + Add custom MCP. On the following page, add the URL of your MCP server to connect your custom MCP. Then give it a name and add your Access token/API key. You'll then see your MCP listed under your tool list.
If an integration you need isn't available, you can build a custom one directly from the agent builder chat. See Build a custom tool with Tools on Demand below.
Brain tab - Skills
Add skills that give your agents the ability to complete specific tasks.
Click on + Add skills. From there, choose a skill from the list that fits your needs, then click Add skill. You can also chat with your agent to create a skill together.
Brain tab - Choose your LLM model
Under Model you can choose the primary AI model for your agent. Additional models may be used by the agent to optimize performance.
Jobs tab
Jobs let you break your agent's work into separate, focused tasks. Each has its own trigger and its own scoped instructions. Instead of one global instruction set, each Job handles a specific piece of work, making your agent more precise and efficient.
Each Job contains:
- A trigger that defines when it runs
- Instructions scoped specifically to that task
Jobs can run in parallel, meaning a complex workflow can be split into multiple Jobs that execute simultaneously. Jobs can also be scheduled to run on a recurring basis: daily, weekly, or monthly.
There are two ways to create Jobs:
- Agent-suggested, via chat: the agent identifies that a task is recurring or complex and proactively suggests creating a Job. You confirm, and the Job is created.
- Manual, from the Jobs tab: go to the Jobs tab and create a Job directly by picking a trigger, writing instructions specific to that Job, and giving it a name.
Two default Jobs exist on every agent: When Assigned and When Mentioned. These are always available regardless of how many additional Jobs you create.
You can ask the agent to perform a one-off task at any time, regardless of its existing Jobs. The agent executes it and returns the result without running its full instruction set.
Channels tab
In the Channels tab, you can define how your agent communicates with people in external apps. These apps include:
- Telegram
With this feature, you can directly activate and also communicate with your agent from outside the monday platform using the communication tool you prefer.
For example, if you're on the go and prefer communicating on WhatsApp, connect your agent to WhatsApp and message it directly on your WhatsApp mobile app.
Build a custom tool with Tools on Demand
If an integration you need is not available inside the platform, you can build one directly from the agent builder chat. Describe the external service and the actions you need in plain language, and the tool will be researched, generated, and deployed to your account automatically.
Once built, the tool is available to any agent on your account. You can find it under the Created by Me tab when adding tools to an agent.
To build a tool, type your request into the agent builder chat. For example:
- "I need to fetch open issues from Linear"
- "Build a Stripe integration to create and list payments"
- "I want to manage contacts in HubSpot"
If no existing integration matches your request, the agent will offer to build one. Click Yes, build it to start. The build runs in the background and typically takes a few minutes. You can continue working on your agent while it runs.
When the build completes, a connection card appears in the chat. Click Connect and enter your API key or token for the external service. You can find your API key in the Settings or Developer section of the external service.
Manage skills from the Skill Hub
On the agent builder home page, you can view any pre-existing agents. Add new capabilities to existing agents by visiting the Skill Hub:
Browse through existing skills, upload skills, or create new ones by clicking on Create skills:
After you click on Create skill, you will be able to chat with monday sidekick and describe exactly which skill you want your agent to perform and customize it as you wish. You can also manually write out what you'd like to create. All changes are saved automatically.
If you have any questions, please reach out to our team right here. We’re available 24/7 and happy to help.