chore: add hitl for generic agent

Author: radugheo

samples/generic-rag-agent/main.py

@@ -8,6 +8,8 @@
 from llama_index.core.workflow import (
     Context,
     Event,
+    HumanResponseEvent,
+    InputRequiredEvent,
     StartEvent,
     StopEvent,
     Workflow,
@@ -42,8 +44,8 @@
 uipath = UiPath()
 
 class CustomStartEvent(StartEvent):
-    query: str
-    add_data_to_index: bool
+    query: str = ""
+    add_data_to_index: bool = False
 
 
 class QueryEvent(Event):
@@ -66,6 +68,12 @@ class AgentAnswerEvent(Event):
     agent_answer: str
 
 
+class FormattedAnswerEvent(Event):
+    """Event containing the formatted answer awaiting human confirmation."""
+    formatted_answer: str
+    iteration_count: int
+
+
 class OutputEvent(StopEvent):
     """Event representing the final, formatted output."""
     output: str
@@ -154,6 +162,8 @@ async def workflow_entrypoint(
         self, ctx: Context, ev: CustomStartEvent
     ) -> QueryEvent | AddDataToIndexEvent:
         await ctx.store.set("original_query", ev.query)
+        await ctx.store.set("iteration_count", 0)
+        await ctx.store.set("feedback_history", [])
 
         if ev.add_data_to_index:
             return AddDataToIndexEvent()
@@ -322,13 +332,22 @@ async def process_query(
     @step
     async def format_final_answer(
         self, ctx: Context, ev: AgentAnswerEvent
-    ) -> OutputEvent:
+    ) -> FormattedAnswerEvent:
         """
         Takes the ReAct Agent's raw answer and formats it into the final structure.
         """
         original_query = await ctx.store.get("original_query")
+        iteration_count = await ctx.store.get("iteration_count")
+        feedback_history = await ctx.store.get("feedback_history", [])
         agent_answer = ev.agent_answer
 
+        feedback_context = ""
+        if feedback_history:
+            feedback_context = "\n\nPrevious feedback from the user:\n"
+            for i, feedback in enumerate(feedback_history, 1):
+                feedback_context += f"{i}. {feedback}\n"
+            feedback_context += "\nPlease address the feedback above in your reformatted answer.\n"
+
         prompt = f"""
             You are a final synthesis engine. You are given the user's original query and
             a comprehensive answer generated by an intelligent multi-tool agent.
@@ -352,6 +371,7 @@ async def format_final_answer(
             If applicable, suggest actionable items or state relevant policies.
 
             Original query: {original_query}
+            {feedback_context}
 
             Agent's Comprehensive Answer to be Formatted:
             ---
@@ -364,10 +384,59 @@ async def format_final_answer(
 
         print("Final formatting complete.")
 
-        return OutputEvent(
-            output=response.text,
+        return FormattedAnswerEvent(
+            formatted_answer=response.text,
+            iteration_count=iteration_count
+        )
+
+    @step
+    async def request_human_confirmation(
+        self, ctx: Context, ev: FormattedAnswerEvent
+    ) -> OutputEvent | QueryEvent:
+        """
+        Present the formatted answer to the user and request confirmation.
+        """
+        await ctx.store.set("last_formatted_answer", ev.formatted_answer)
+
+        print("\n" + "="*80)
+        print("FORMATTED ANSWER:")
+        print("="*80)
+        print(ev.formatted_answer)
+        print("="*80)
+
+        ctx.write_event_to_stream(
+            InputRequiredEvent(
+                prefix="\nIs this answer satisfactory? (yes/no or provide feedback): "
+            )
         )
 
+        response = await ctx.wait_for_event(HumanResponseEvent)
+        feedback = response.response.strip().lower()
+
+        print(f"Received response: {feedback}")
+
+        if feedback == "yes":
+            print("\nAnswer approved by user.")
+            return OutputEvent(output=ev.formatted_answer)
+
+        max_iterations = 3
+        iteration_count = ev.iteration_count
+
+        if iteration_count >= max_iterations:
+            print(f"\nMaximum iterations ({max_iterations}) reached.")
+            return OutputEvent(
+                output=f"{ev.formatted_answer}\n\n---\nNote: Maximum iteration limit reached. Please refine your query or requirements."
+            )
+
+        feedback_history = await ctx.store.get("feedback_history", [])
+        feedback_history.append(feedback)
+        await ctx.store.set("feedback_history", feedback_history)
+        await ctx.store.set("iteration_count", iteration_count + 1)
+
+        print(f"\nFeedback received: '{feedback}'. Re-querying (iteration {iteration_count + 1})...")
+
+        return QueryEvent()
+
 
 agent = AgentWorkflow(timeout=600, verbose=True)
 

samples/generic-rag-agent/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "generic-rag-agent"
-version = "0.0.24"
+version = "0.0.27"
 description = "A generic RAG agent that combines multiple knowledge bases with web search capabilities and MCP tool support"
 authors = [{ name = "John Doe", email = "john.doe@myemail.com" }]
 readme = { file = "README.md", content-type = "text/markdown" }