Service AI agents are part of service AI in monday service. They let you build AI teams that can answer requests, route conversations to the right agent, and hand them over to a human expert when needed.
In this article, you'll learn how to create a Service AI agent team, configure agents, manage knowledge sources, connect channels and portals, test the experience, and monitor performance.
What are Service AI agents?
Service AI agents are the AI teammates you create and manage across your organization.
Instead of relying on a single general agent, service AI uses multiple specialized agent teams. Each agent is set up for a specific type of request, such as benefits, payroll, onboarding, or IT support. This helps improve routing accuracy and makes the setup easier to maintain over time.
When a request comes in, the Service AI Supervisor reviews the request and routes it to the most relevant agent based on that agent's configured scope. If AI cannot confidently handle the request, the conversation is routed to a human expert.
Keep reading for the steps to configure your agents.
Open Service AI
1Open monday service.
The left pane includes Requests, where you can access your workspaces and Tickets boards, alongside the tabs that make up Service AI:
- Agents - build, manage, and test your AI teams
- Performance - monitor how your Service AI agents are doing
- Portals - manage where requesters go for self-service and AI chat
- Knowledge - manage docs and connected knowledge sources
- Channels - manage how requests come in
2Click Agents to get started.
Create a team
Start by creating a new team:
Choose the team type that best fits the requests you want the team to handle, and the requests workspace you want your team to collaborate in. The requests workspace needs at least one Tickets board linked to the team:
Your new team appears on Manage agents alongside your other teams. If you chose a department type with presets, the team includes ready-made agents for that department. For example, an IT team may include agents such as IT Support, Access & Identity, and Device & Asset agents. Each agent's card shows its scope description and status (Active or Inactive). Once an agent starts resolving requests, its resolution rate appears next to the status.
Click +New agent on a team card to create additional agents beyond the presets. Use the three-dot menu next to it to add another agent, open team settings, or delete the team. Team settings let you change the connected requests workspace and view or add incoming request channels.
Next to +New agent, icons show which channels are connected to the team's requests (Portal AI chat, email, and forms). Hover over an icon to see its connection status.
Configure an agent
Click into any agent to open its setup panel. Each agent has five tabs: Profile, Knowledge, Playbook & Skills, Coverage Test, and Activity. You can also activate or deactivate, or delete the agent from the top of the panel. An agent must be active in order to answer requests. To activate an agent, it must have both scope and knowledge.
Profile
In the Profile tab, define the agent's mission and scope, tone of voice (natural, professional, or friendly), and answer length. The Mission & Scope field is the main thing that affects routing. The Service AI Supervisor uses it to decide which requests to route to that agent. It does not use the knowledge base to make that routing decision, so your scope should be as specific as possible.
A narrow scope like vacation requests, PTO, and leave policies is more reliable than a broad scope like HR questions. In the same way, hardware, software, and connectivity requests is more useful than IT stuff. Be careful not to include capabilities your team does not actually support. For example, if you do not use device-management tools, avoid promising asset or device-management help in the agent's scope.
Knowledge
In the Knowledge tab, add the information the agent uses to answer requests. Knowledge is split into two types: Sourced knowledge, for content you want requesters to see, where the agent cites the source and includes a link when it answers, and Unsourced knowledge, for content shared without its source shown to the requester.
Click Add sourced knowledge to add a source from:
- monday docs: pull in a monday doc.
- Add files: upload a file directly. See file limitations below.
- Web page: sync a public web page. The agent syncs the page and one level of linked sub-pages, and refreshes weekly. Pages that require login aren't supported.
- Confluence Cloud Spaces: pull from a Confluence Cloud space your admin has connected under Knowledge > Connectors. You can select specific spaces, or a specific hierarchy of articles within a space, for this agent. Content syncs from Confluence Cloud once a day.
- SharePoint Sites: pull from a SharePoint site your admin has connected under Knowledge > Connectors. You can select specific sites, or a specific folder hierarchy within a site, for this agent. Content syncs from SharePoint once a day.
Click Add knowledge under Unsourced knowledge to add a source from the same list, plus one more:
- monday board: pull from any board across your account, except Tickets boards. File columns aren't extracted, and board content isn't matched against personal information, so use caution when adding boards that contain sensitive data.
File upload limitations
| File type | Supported languages | Size limit | Other limits |
| Documents: txt, md, html, xls, xlsx | 100+ languages | 25MB each | Up to 3,000 pages per document |
| Documents: doc, docx, pdf | English, German, Spanish, French, Italian, Portuguese | 25MB each | Up to 3,000 pages per document |
| Images: tiff, jpeg, png | English, German, Spanish, French, Italian, Portuguese | 5MB each | Up to 8K resolution (7,680 x 4,320 pixels) |
Password-protected files can't be uploaded, regardless of type.
Playbook & Skills
In the Playbook & Skills tab, review the built-in playbooks that shape agent behavior: Route to human, Content safety & language, Confidentiality, Response behavior, and Staying on-topic / out-of-scope. These are Core playbooks owned by Service AI. Editing existing playbooks and adding new ones is coming soon.
Coverage Test
In the Coverage Test tab, review auto-generated sample requests based on the agent's scope, knowledge, and skills. This helps you understand whether the agent is likely to resolve the request, partially cover it, or route it to a human. The tab shows a knowledge coverage level, such as Intermediate, based on a set of auto-generated tests, along with a breakdown of how many were answered directly versus routed to a human expert. Each sample question includes the agent's reasoning, such as whether the request was in scope and whether knowledge was sufficient.
Click Regenerate for a new set of sample questions, or Test within team to test the agent alongside the rest of its team.
Activity
Once the agent is live and handling requests, use the Activity tab to review its activity log: what happened (such as agent replied, ticket resolved, or handed off to human), the related ticket, the channel the request came through, the reasoning behind the action (such as "Answered with KB" or "Matched - high confidence"), and when it happened. Filter the log by event type or time range (last 7 or 30 days).
Service AI Supervisor
The Service AI Supervisor helps you understand how requests are matched and routed across your Service AI agents. In the left pane, alongside Manage agents and Test, you'll find two collapsible sections: Service Frontliners, listing all the agent teams you create, and Service Ops, where the Supervisor and Service Sidekick live:
When a request comes in, the supervisor recognizes the request intent and checks whether there is a matching AI agent based on that agent’s configured scope. If the supervisor finds a match, the conversation is assigned to that AI agent. If no matching AI agent is found, the outcome depends on the channel. In portal chat (we'll cover it later in this section), the supervisor can first try to help using the portal’s resources before routing the request to a human expert. In other channels, the request is routed directly to a human.
Before an AI agent answers, the system checks whether the agent has enough knowledge coverage for the request. If knowledge coverage is too low, the request is routed to a human instead of being answered by AI.
Click Supervisor under Service Ops to review its activity and see how your service flow is set up.
Activity
The Activity tab shows how well requests are being matched for a selected workspace: Matched requests, Unmatched requests, and Match rate. Below that, the activity feed lists each routing decision: the event (such as "Assigned to Agent"), the related ticket, the channel the request came through, the reasoning behind the routing decision (such as "Matched, high confidence"), and when it happened. Filter the feed by workspace, event type, or time range.
Flow
The Flow tab shows a visual map of how your service handles requests for a selected workspace: the channels requests come in through, the board the request is created on, and how the Supervisor routes it. Requests either match an agent's scope and go to In scope, where they're assigned to the matching agent, or fall under No match, where they're assigned to a human expert instead.
Knowledge
Click Knowledge in the left pane to manage the content your Service AI agents can draw from. This section has two tabs: Docs hub and Connectors.
Docs hub
Docs hub is where you create and browse monday docs. Click Create new doc to start from scratch, or Create doc with AI to generate one. You can also start from a template, such as Meeting Notes or Employee Onboarding, or open a recent document.
Docs created here can be added as Sourced knowledge on any agent's Knowledge tab.
Connectors
Connectors is where an admin links third-party platforms to your account so agents can use their content as knowledge. Click Add connector, and choose a platform:
- Confluence Cloud: sync spaces.
- SharePoint: sync sites.
Choosing a platform opens its login screen, Atlassian for Confluence Cloud, or Microsoft for SharePoint, so the admin can sign in and authorize the connection.
After signing in, select which sites or spaces to sync. This includes all nested files within each selection, except unsupported formats. Click Add to complete the connection.
Once added, the connector shows as Processing while it syncs. This usually takes a few minutes, and agents won't have access to the content until syncing completes. Each synced site or space is listed with its file count, and you can open it directly in the source platform.
Once syncing completes, the connector shows its total file count and last sync date. Content syncs once a day, so any changes made in Confluence Cloud or SharePoint appear here within a day.
Use the connector's three-dot menu to open the source platform directly, sync it manually, edit which sites or spaces are included, or disconnect it.
Once a platform is connected, individual agents can select specific spaces or sites, or a more granular hierarchy within them, from their own Knowledge tab, as covered in the Configure an agent section above. When an agent answers using this content, it cites the source and links directly back to the page in Confluence Cloud or SharePoint.
Connect channels
Click Channels in the left pane to bring all your channels into one place and manage how requests enter your service workflows.
From this tab, you can add a new channel, search existing channels, and filter the list by channel type or owner. Each channel row shows the channel name, the Tickets board it creates tickets in, the owner, and whether it's connected.
To add a channel, click Add a channel and choose a type:
Each channel is connected to a Tickets board and linked to a workspace. Once a channel is connected to a board inside the team's workspace, that team's agents can work on requests from that channel, including email channels. If there is no matching agent, the request is routed to a human.
To edit a channel, click its edit icon, or open its three-dot menu for more options. Editing a channel lets you update its name, the connected Tickets board, and its email address:
Who can edit channels
On the Channels tab, anyone can view existing channel connections and create a new connection, whether the workspace is open or closed. However, only the team member who created a connection can edit that specific connection.
This means workspace type does not affect who can view or create channel connections. Editing permissions stay with the original creator of each connection.
Manage portals
The Portals page brings all your portals into one place.
If your account has one portal, this page shows that portal marked as a Default portal.
If you are on the Enterprise plan and use multiple portals, this page shows your main portal as well as your additional portals. To create another portal in a multi-portal setup, click Create portal.
To create a portal, name it, set its URL slug, and choose a workspace and ticket board:
Search or filter existing portals by owner, access type, or workspace. Each row shows the portal's request channels, its Service AI agents team, and its owners.
From the Portals page, you can open a portal in two ways:
- Click the portal itself to open its settings.
- Click the three-dot menu and choose Edit portal to open settings, Go to portal to open the live portal.
In a portal's settings, manage its Content, AI Chat, Access, and Users, or click Open visual editor to customize its design and layout.
Who can edit portals
Permissions depend on whether you're working with the main portal or a workspace portal.
For the main portal, anyone can view and edit the Content, AI Chat, and visual editor. Only account admins can edit the Access and Users tabs.
For workspace portals, permissions depend on the workspace type, whether the portal is open or private, and whether the person is a workspace member or not.
If someone is a workspace member, an open portal can be seen by anyone, and anyone can access its settings. A private portal can only be seen by portal subscribers and admins, and only they can access its settings.
If someone is not a workspace member in a Closed workspace, both open and private portals can only be seen by portal subscribers and admins, and only they can access portal settings.
Set up AI chat in your portal
Open a portal's settings and select the AI Chat tab.
Connect AI chat to a Tickets board. This step is required because tickets created through AI chat are added to the selected board, and that board connects the portal experience to the Service AI agents team.
Below the ticket board, the Workforce agents section shows whether an agent team is active for this portal. If none is connected yet, click Create AI workforce to set one up. Once a team is active, this section shows how many agents are active, which team they belong to, and lets you Manage that team directly. Below that, each individual agent and its status is listed.
Once connected, the tab shows Connected status, and a Test AI chat link appears so you can try the experience directly before requesters do.
Once AI chat is connected, requesters can open the portal and click New ticket. They can then choose how to open a ticket: Get assistance from AI, or one of the available ticket forms, grouped by category.
Choosing Get assistance from AI opens a chat where the requester can type their question:
As soon as the conversation starts, a ticket is created in the connected Tickets board. The Service AI Supervisor then reviews the request and checks whether there is a matching AI agent for that request type. If there is, the request is routed to that agent. If the agent has enough knowledge coverage, it responds in the portal. If it cannot handle the request, the conversation is routed to a human expert.
If no matching agent is found in portal chat, the supervisor can first try to help using the portal’s available resources before routing the request to a human. In that case, the supervisor uses the portal resource titles and descriptions, not the full content of those resources.
When a request is created through portal AI chat, it appears in the connected Tickets board.
The board includes a dedicated AI Status column that tracks where the request is in the AI flow. The current AI status values are In progress, Pending customer response, AI resolved, and Routed to human.
The AI Status column and your regular workflow Status column are separate. AI Status is system-set and cannot be edited manually. Your regular Status column controls your board workflow and automations.
When AI resolves a request
A request is marked as AI resolved only after the system determines that AI handled it successfully without routing it to a human. This is not decided immediately.
The system waits 24 hours after the last AI response in portal conversations and 72 hours after the last AI response in email conversations before marking the request as resolved.
After a ticket is marked AI resolved, any new requester reply routes the ticket to a human expert. This applies in both portal and email. The AI Status changes to Routed to human, and you can use automations so your regular workflow status reflects that return as well.
Test the full workforce flow
Click Agents in the left pane, then click Test, located below Manage agents, to simulate the requester experience across your Service AI agents:
Select a workspace. Portal AI chat is currently the only channel type available for testing; Email and Forms are coming soon.
Start a conversation to see how the Supervisor routes the request and how the flow behaves end to end:
As covered earlier, the left pane also lists your Service Frontliners and Service Ops, so you can jump into any agent's setup or the Supervisor's activity without leaving the test flow.
Each test is saved as its own conversation, so you can return to previous tests, compare outcomes, and continue reviewing them over time. You can also create new tests and delete tests you no longer need.
This is different from the Coverage Test tab inside a single agent. The full Test page is designed for testing across your whole Service AI setup, while Coverage Test is designed for agent-level testing.
Review AI workforce performance
Click Performance in the left pane to monitor how your Service AI agents are performing over time. Filter by date range and workspace.
The page opens with three summary cards: AI resolved requests, AI involvement rate, and AI resolution rate, each showing the current figure alongside how much it's changed. The Performance funnel shows how requests move through the flow: total requests, how many involved AI, and the split between AI resolved, in progress, no match, and routed to human.
The AI agent performance table lists every agent, its team, request volume, and resolution rate, so you can compare performance across your whole setup. The Documentation performance table shows which knowledge sources are driving resolutions: how often each document was used in a request, how many of those led to an AI resolution, and its resolution rate. Click Improve doc next to a document to strengthen it.
Use this page to understand how often AI is involved, how often it resolves requests successfully, and how often requests still need a human handoff.
Requests dashboards
Below Performance, Requests dashboards links to monday dashboards built on your ticket data, such as an IT Analytics Dashboard, Reporting Dashboard, and any Tickets dashboards set up on your account.
If you have any questions, please reach out to our team right here. We’re available 24/7 and happy to help.