UiPath Documentation
agents
latest
false
Agents user guide

Deploying conversational agents

Publish and deploy a conversational agent to Orchestrator to make it available to users after design and testing are complete.

After designing and testing your conversational agent, you need to publish and deploy it to make it available to users. Publishing and deployment are two separate steps:

  • Publishing: Packages your agent as a solution and uploads it to Orchestrator.
  • Deployment: Makes the published solution available for execution with the required runtime resources.

Orchestrator folder requirements

Before publishing, ensure your target Orchestrator folder has the required resources configured.

ResourcePurposeRequired
Serverless robot machine templateExecutes the conversational agent and specific tool calls (cross-platform RPA automations)Always
Unattended robot accountProvides an execution identity when agent is used anonymouslyIf the agent is accessed through a channel using anonymous access
Virtual machineExecutes Windows-based RPA automationsIf using Windows automations
Shared connectionsProvides access to Integration Service connectors when no user-configured personal connection is availableIf tools use connectors
Important:

If your agent fails to start in Instance Management, verify that:

  • You are in the correct tenant where the agent solution resides.
  • A serverless robot machine template is assigned to the solution folder.
  • If using an anonymous-access channel, an unattended robot account is assigned to the solution folder.

Execution identity

  • When the conversational agent is accessed through a deployment channel that authenticates with the user's UiPath account, the conversational agent runs under the user's identity. When the agent invokes tools that are Orchestrator processes, they run under the user's identity as well.
  • When using a deployment channel with anonymous access, the conversational agent instead runs under the identity of an unattended robot account within the folder.

Publishing

Publishing packages your conversational agent as a solution and uploads it to Orchestrator.

Publishing steps

  1. In Studio Web, open your conversational agent.
  2. Select Publish in the top toolbar.
  3. Select the target Orchestrator feed (we recommend Orchestrator tenant feed).
  4. (Optional) Enter a Change log note describing your change.
  5. Select Publish.

For detailed publishing options, see Publishing a project.

Deployment

After publishing, you need to deploy the solution to make it available for execution. Deployment associates the published solution with the runtime resources in your Orchestrator folder.

Deployment steps

  1. Navigate to your Orchestrator.
  2. Locate the published solution.
  3. Deploy (or upgrade) the solution, ensuring it has access to the required robot accounts and connections.

Once deployed, the agent appears in Instance Management and becomes available through configured channels.

For detailed deployment options, refer to Publishing a project.

Connection management

The agent needs access to the same connections used at design time. Conversational agents support two types of connections: shared connections configured by admins, and personal connections configured by individual users.

Shared connections

Shared connections are configured by admins in an Orchestrator folder and are available to all users of the agent. To set up shared connections:

  1. Create Integration Service connections in a shared Orchestrator folder.
  2. Use these shared connections during design and debugging.
  3. Publish the solution as a subfolder within the same shared folder.

This setup ensures the solution automatically inherits the parent folder's connections.

Personal connections

Personal connections allow individual users to select which Integration Service connections the agent uses on their behalf. This is useful when users need the agent to act with their own credentials — for example, sending emails from their own account or accessing their personal calendar.

Enabling personal connections

For personal connections to work, two settings must be configured on each tool process in Orchestrator. For general information about execution settings, see Linking execution settings.

  1. In Orchestrator, navigate to the folder where the agent solution is deployed and open the Processes tab.
  2. Select Edit on the tool process that uses a connector (for example, an API workflow that sends emails).
  3. In the process configuration, set the Account to Inherit parent job identity. This allows the tool to run under the user's identity, which is required for the tool to access the user's personal connections.
  4. Set the connector's connection to Configurable by users. This allows users to select their own connections for the connector at runtime.

Important:

Both settings are required. Without either setting, personal connections are not visible to users and the tool may return an error at runtime.

User experience

When personal connections are enabled for an agent:

  • Users can open the agent's Settings panel and select the Available Connections tab to view and manage their connections.
  • For each required connector (for example, Slack, Outlook, or Salesforce), users can select one of their available connections or create a new one.
  • By default, the agent auto-selects a matching connection from the user's personal workspace for the connector that the agent and other tools in the folder use.

The agent attempts to auto-configure connections from the user's personal workspace. If the user has a matching connection for a connector in their personal workspace, the agent uses it automatically — making setup seamless and typically a one-time experience. If the auto-configured connection fails or no matching connection is found, the user is prompted to select or create a new one.

If a user has not configured a personal connection for a connector, the agent falls back to the agent's default connection. Selected personal connections take precedence over shared connections and are always used when present. When using external applications, only connections from the user's personal workspace are available — shared folder connections are not displayed.

Note:

Users can select and manage personal connections only in channels that support user-profile settings (see Feature availability by channel). Once configured, personal connections are used by the agent across all channels.

To learn how to design an agent for personal connections, see Connection considerations in the design guide.

Deployment channels

After publishing, users can access your conversational agent through multiple channels. Each channel offers different capabilities and is suited for different use cases.

Channel overview

ChannelBest forUser authentication
Instance ManagementAdmin testing, internal usersUiPath account
Autopilot for EveryoneInternal productivity hubUiPath account
Microsoft TeamsTeam collaborationMicrosoft account and UiPath account
SlackTeam collaborationSlack account and UiPath account
iFrame embeddingThird-party apps and UiPath AppsUiPath account or anonymous
UiPath TypeScript SDKCustom-built applicationsDetermined by your application

Feature availability by channel

Not all chat features are available in every channel. Instance Management provides the complete feature set; other channels may have limitations.

FeatureInstance ManagementAutopilot for EveryoneMicrosoft TeamsSlackiFrameTypeScript SDK*
Start new chat
Chat history
Delete chat session
User-profile settings
Personal connections
Starting prompts
File uploads
Conversation inputs
Tool-call confirmations
Citations
HTML preview
Copy response
Feedback (thumbs)

* All SDK features can be configured by the developer in their custom implementation.

Choosing a channel

Consider these factors when selecting deployment channels:

Internal users with UiPath access
  • Use Instance Management for admin testing and power users.
  • Use Autopilot for Everyone as a central hub for all agents.
Team collaboration
  • Use Teams or Slack to bring the agent where teams already work.
Custom applications
  • Use iFrame embedding for third-party apps or UiPath Apps.
External or anonymous users
  • Use iFrame embedding with anonymous auth server.
Custom applications
  • Use the UiPath TypeScript SDK to integrate conversational agents into your own web applications or portals.

Chat session behavior

Conversational agents maintain persistent sessions:

  • Each conversation runs as a dedicated Orchestrator process (active for up to 8 hours of inactivity).
  • Sessions remain active to enable immediate responses.
  • Idle sessions consume minimal resources.

Using UiPath tools in external platforms

You can expose UiPath capabilities (automations, workflows, document processing) to external chat platforms like ChatGPT and Microsoft Copilot Studio using MCP (Model Context Protocol) tools. This approach uses direct tool calls rather than conversational agents.

For details on MCP integration, refer to MCP Servers.

Next steps

Select a channel to learn about setup and configuration:

Was this page helpful?

Connect

Need help? Support

Want to learn? UiPath Academy

Have questions? UiPath Forum

Stay updated