I am working on a UiPath Conversation Agent for Orchestrator-related queries like job status, failure reason, queue count, ETA, and next run schedule.
Currently, the agent can maintain context within the same active session. For example, after asking about a process status, the user can ask “Why did it fail?” and the agent understands the context.
My question is:
Is there any way in UiPath to fine-tune or improve the Conversation Agent/LLM behavior based on older conversations, previous sessions, user feedback, or incorrect responses?
Also, if built-in fine-tuning is not available, what is the recommended approach? Should we store previous conversations separately and use them to improve prompts, tool routing, and evaluation cases?
Currently, UiPath Conversation Agents support session-level context, but there is no built-in fine-tuning based on previous conversations or user feedback.
Recommended approach is to store conversation history (DB/Data Service) and use it with prompts/RAG to improve context and evaluations over time. For Orchestrator query use cases, structured tool outputs + retrieval-based context usually provide more reliable results.
Currently, UiPath Conversation Agents can maintain context within the active session, but there is no built-in feature to continuously train or fine-tune the agent using previous conversations, user feedback, or incorrect responses across sessions.
I strongly recommend you to :
Store conversation history externally
Maintain session and historical context separately
Use previous conversations to improve prompts and tool routing
Build evaluation/test cases from real user interactions
Capture feedback and failed responses for continuous improvement
For Orchestrator-related agents like job status, queue count, schedules, etc., a good approach is to use the LLM mainly for conversational understanding while keeping memory, analytics, and business context outside the model.