# Limitations and FAQ > This page documents current limitations of conversational agents and provides answers to frequently asked questions. This page documents current limitations of conversational agents and provides answers to frequently asked questions. ## Current limitations We are actively expanding the product with new capabilities. The table below outlines the current limitations, which are planned to be addressed in upcoming releases. | Limitation | Description | |------------|-------------| | **Local desktop automation** | The agent cannot run automations on the user's local desktop via Assistant. | | **Personal connections** | Only shared Integration Service connections are available. Personal connections are not supported. | | **Tool confirmation** | The agent does not ask for user confirmation before executing tools. | | **PII masking** | PII detection and masking is not available. | | **Prompt injection guardrails** | Agent-level guardrails for prompt injection are not available. | | **File size** | File uploads are limited to 5MB. | | **Voice interaction** | Text-only interaction. Push-to-talk and two-way voice are not available. | | **IVR integration** | Interactive voice response integration is not available. | ### Channel feature availability Not all features are available in every deployment channel. Refer to [Deployment](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/conversational-agents-deployment) for the full feature comparison. Key differences: | Feature | Where it works | Where it doesn't work | |---------|----------------|----------------------| | Starting prompts | Instance Management, iFrame | Autopilot for Everyone, Teams, Slack | | HTML preview | Instance Management, iFrame | Autopilot for Everyone, Teams, Slack | | File uploads | Instance Management, Autopilot for Everyone, iFrame, Teams, Slack, TypeScript SDK | — | | Citations | Instance Management, Autopilot for Everyone, Teams, Slack, TypeScript SDK | iFrame* | | Debug dump | Instance Management, Autopilot for Everyone | Teams, Slack, iFrame | Additional channel limitations: * **Teams and Slack**: Only direct messages are supported. Channel discussions and @mentions are not available. * **Citation preview**: Only available for PDF documents. * **iFrame citations**: The citation viewer does not work when the iFrame is embedded within a UiPath App. ## FAQ ### Conversational agents overview #### How do conversational agents differ from autonomous agents? Conversational agents are designed for multi-turn, back-and-forth dialogue with users. Autonomous agents execute tasks from a single prompt. Use conversational agents when your scenario requires ongoing clarification, personalized guidance, or real-time interaction. See [Conversational agents](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/conversational-agents) for a detailed comparison. #### How do conversational agents relate to Autopilot for Everyone? They work side-by-side: * **Autopilot for Everyone**: UiPath's general-purpose assistant for productivity tasks. * **Conversational agents**: Specialized agents you build for specific use cases. You can access your conversational agents from within Autopilot for Everyone, making it a central hub for all your conversational needs. ### Licensing and consumption #### How are free actions counted? Free actions are counted per user, not pooled across users. Each user has their own individual free allotment based on their license type: * **Express**: 50 free actions per year * **Basic**: 50 free actions per month * **Plus/Pro**: Unlimited (subject to fair use) If User A sends 20 messages and User B sends 80 messages, User A still has 30 free messages remaining while User B has exceeded their free tier. #### Do tool calls cost extra? Some tools have additional consumption beyond the base action cost: * **No additional cost**: Context Grounding (semantic search), Integration Service connectors, MCP tools Refer to [Licensing](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/conversational-agents-licensing) for complete details. #### What counts as an action? One action equals one user message sent to the conversational agent. It doesn't matter how many agent responses are generated from that message. Messages with eight characters or fewer that don't execute a tool are not counted. ### Models and configuration #### Can I use my own LLM? Yes. Conversational agents support [LLM Configurations](https://docs.uipath.com/automation-cloud/automation-cloud/latest/admin-guide/managing-ai-trust-layer#configuring-llms) so you can use your own LLM subscriptions. This feature is in preview. During the preview period, there is no charge per message, but additional consumption for certain tool usage still applies. #### What file types can users upload? You must add a file analysis tool (Analyze files, IXP, or custom workflow) to enable file uploads. Supported types: * **Images**: GIF, JPE, JPEG, PNG, WEBP * **Documents**: PDF (under 5MB) ### Deployment and access #### Why is my conversational agent failing to start in Instance Management? Check the following: 1. Ensure an unattended robot account is assigned to the solution folder. 2. Verify you are in the correct tenant where the agent solution resides. 3. Check that a serverless robot is assigned if you're using cross-platform RPA tools. #### Why am I getting a connection error when running tools? The agent needs access to the same connections used at design time: 1. Create Integration Service connections in a shared Orchestrator folder. 2. Use the shared connections during design and debugging. 3. Publish the solution as a subfolder within the same shared folder. This ensures the solution automatically inherits the parent folder's connections. #### How can I limit which users can access a specific conversational agent? User access is controlled through Orchestrator folder permissions. Users need appropriate permissions on the folder containing the agent solution. Configure folder access in the Admin portal. ### Embedding and integration #### How can I embed conversational agents in my application? Embed using an iFrame. Refer to [Apps embedding](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/conversational-agents-apps-embedding) for URL construction and parameters. #### Can I embed an agent for users without UiPath accounts? Yes. Use [anonymous access](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/conversational-agents-anonymous-access) to embed agents for external users. This requires setting up an external application and token endpoint. Anonymous users consume from the tenant's unit pool with no free tier. ### Channel-specific issues #### Can I access conversational agents via API? Yes. Conversational agents are available through the [UiPath TypeScript SDK](https://uipath.github.io/uipath-typescript/conversational-agent/). See [UiPath TypeScript SDK](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/conversational-agents-typescript-sdk) for feature availability. #### Why aren't starting prompts showing in Teams, Slack, or Autopilot for Everyone? Starting prompts are only supported in Instance Management and iFrame embedding. Other channels don't propagate these prompts due to platform limitations. #### Can the agent respond in Slack channels or Teams channels? No. Conversational agents only support direct messages through the UiPath Autopilot plug-in. Channel discussions and @mentions are not supported. #### Why can't I see HTML preview in Teams or Slack? HTML preview is only available in Instance Management and iFrame embedding. In other channels, HTML content displays as raw text or markdown. #### Why can't I create conversations in Teams? If conversations fail to start: 1. Verify the agent is published and deployed in Orchestrator. 2. Check that you've connected the Teams app to your UiPath account. 3. Use the **Logout** command and reconnect to refresh your session. 4. Verify the agent is available using the **Switch Agent** command. ### Context Grounding #### Why can't the agent find my Context Grounding index? If you receive "index not found" errors: 1. Verify the index exists and is in a "Ready" state. 2. Check that the agent has access to the folder containing the index. 3. Ensure the index hasn't been moved or deleted since configuration. 4. Republish the agent after making any index changes. #### Why are citations incorrect or missing? Citation quality depends on: * **Index quality**: Ensure source documents are well-structured * **Chunk size**: Very large or small chunks affect citation accuracy * **Query relevance**: Ambiguous queries may retrieve less relevant content Note: Citation preview is only available for PDF documents. ### Performance #### What factors affect response time? Several factors influence latency: * **Model selection**: Different models have different latency characteristics * **Context window size**: Larger conversation histories increase processing time * **Tool execution**: RPA workflows and API calls add execution time * **Context Grounding**: RAG searches add latency based on index size To optimize: * Use API workflows instead of RPA for API-only operations. * Keep conversation context focused. * Choose appropriate models for your latency requirements. ### Compliance and governance #### How can I audit LLM calls and agent behavior? The AI Trust Layer provides a complete audit trail. Access audit logs through the Admin portal for: * Complete history of model interactions. * Token usage tracking. * Policy compliance verification. See [Observability](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/conversational-agents-observability) for details. ### Data storage #### Where is chat history stored? Chat history is stored in Azure database, scoped to the tenant. Users can only access chat history for agents they have permission to use. #### Where are uploaded files stored? Uploaded files are stored in the associated tenant's system storage bucket. This storage is not accessible from the Orchestrator UI. #### How long are uploaded files retained? Files follow Orchestrator's retention policy. The retention timer does not start until the job completes. The default retention period is 30 days for Community and Enterprise editions, and is configurable up to 6 months in Orchestrator settings. ### Customization #### Can I customize the welcome message? When embedding via iFrame, use the `welcomeTitle`, `welcomeDescription`, and `suggestions` URL parameters. See [Apps embedding](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/conversational-agents-apps-embedding) for details. In Instance Management, the welcome experience uses the agent's configured settings. ## Getting help If you encounter issues not covered here: * Review [Agent traces](https://docs.uipath.com/agents/automation-cloud/latest/user-guide/agent-traces) for debugging information. * Use the Debug dump feature (Ctrl+Alt+D in Instance Management) to capture diagnostic information. * Contact UiPath support with the debug dump and relevant trace information.