feat: implement RAG capabilities and Context Pane integration

- Add RAG capabilities using LlamaIndex and ChromaDB
- Implement RAGManager for PHB indexing and retrieval
- Integrate RAG pipeline into orchestrator to trigger queries based on extracted entities
- Update TUI to include a 3-column layout with a real-time Context Pane
- Define ContextUpdate data models in src/llm/models.py
- Update requirements.txt with new dependencies

src/pipeline/orchestrator.py

@@ -6,8 +6,9 @@ from typing import List, Optional
 
 import numpy as np
 
-from src.llm.models import ExtractionResult
+from src.llm.models import ContextUpdate, ExtractionResult
 from src.llm.processor import LLMProcessor
+from src.rag.manager import RAGManager
 from src.stt.listener import AudioListener
 from src.stt.transcriber import Transcriber
 from src.ui.tui import ConfirmationApp
@@ -37,10 +38,12 @@ class PipelineOrchestrator:
         self.listener = AudioListener(loop=self.loop)
         self.transcriber = Transcriber(model_size="small")
         self.processor = LLMProcessor()
+        self.rag_manager = RAGManager()
 
         # Queues
         self.transcript_queue = asyncio.Queue()
         self.proposal_queue = asyncio.Queue()
+        self.context_queue = asyncio.Queue()
 
         self.is_running = False
 
@@ -148,6 +151,9 @@ class PipelineOrchestrator:
                         f"LLM Worker: Proposal generated. Putting into proposal queue. (Lore: {len(result.lore_updates)}, Char: {len(result.character_updates)})"
                     )
                     await self.proposal_queue.put(result)
+
+                    # Trigger RAG query based on extracted entities
+                    await self._trigger_rag_queries(result)
                 else:
                     logger.info("LLM Worker: No relevant game data extracted.")
 
@@ -157,6 +163,43 @@ class PipelineOrchestrator:
             # Small sleep
             await asyncio.sleep(0.1)
 
+    async def _trigger_rag_queries(self, result: ExtractionResult):
+        """
+        Triggers RAG queries based on the extracted entities and results.
+        """
+        queries = set()
+
+        # Collect entities from lore updates
+        for update in result.lore_updates:
+            if update.entity_name:
+                queries.add(update.entity_name)
+
+        # Collect entities from character updates
+        for update in result.character_updates:
+            if update.character_name:
+                queries.add(update.character_name)
+
+        # Collect events as potential queries
+        for event in result.significant_events:
+            queries.add(event)
+
+        if not queries:
+            logger.info("RAG: No query terms identified from extraction result.")
+            return
+
+        for query in queries:
+            logger.info(f"RAG: Triggering query for: {query}")
+            try:
+                # Run retrieval in a thread to avoid blocking the event loop
+                updates = await asyncio.to_thread(self.rag_manager.retrieve, query)
+                for update in updates:
+                    await self.context_queue.put(update)
+                    logger.info(
+                        f"RAG: Retrieved snippet for {query} from {update.source}"
+                    )
+            except Exception as e:
+                logger.error(f"RAG: Error retrieving context for {query}: {e}")
+
     def _get_wiki_context(self) -> str:
         """
         Reads all files in the lore directory and returns them as a 저희 context string.
@@ -188,8 +231,11 @@ class PipelineOrchestrator:
         logger.info("TUI Worker started.")
         try:
             # Launch TUI exactly once.
-            # Pass the proposal queue to the app.
-            app = ConfirmationApp(proposal_queue=self.proposal_queue)
+            # Pass the proposal queue and context queue to the app.
+            app = ConfirmationApp(
+                proposal_queue=self.proposal_queue,
+                context_queue=self.context_queue,
+            )
             await app.run_async()
             self.stop()
         except Exception as e: