fixing kg

app.py

@@ -314,16 +314,60 @@ class AbaloneRAGApp:
                 st.error("Model not initialized. Please wait for the knowledge base and model to be ready or rebuild the vectorstore.")
                 return
 
-            if getattr(self, 'use_kg', False):
-                result = answer_with_kg(
-                    self.chain,
-                    styled_question,
-                    prior_history,
-                    persist_dir=self.persist_dir,
-                    kg_hops=self.kg_hops,
-                )
-            else:
-                result = self.chain({"question": styled_question, "chat_history": prior_history})
+            # Call the chain with a safe retry: if the underlying vectorstore is corrupted or missing
+            # (for example, Chroma raises an internal HNSW/disk error), attempt one automatic rebuild
+            # and retry. This avoids crashing the Streamlit app in deployed environments.
+            attempted_rebuild = False
+            last_exception = None
+            while True:
+                try:
+                    if getattr(self, 'use_kg', False):
+                        result = answer_with_kg(
+                            self.chain,
+                            styled_question,
+                            prior_history,
+                            persist_dir=self.persist_dir,
+                            kg_hops=self.kg_hops,
+                        )
+                    else:
+                        result = self.chain({"question": styled_question, "chat_history": prior_history})
+                    break
+                except Exception as e:
+                    # Keep the exception for logging and potential re-raise after a failed retry
+                    last_exception = e
+                    # If we've already attempted a rebuild, give up and show an error
+                    if attempted_rebuild:
+                        st.error("Retrieval error: failed to query the knowledge base. Try rebuilding the vectorstore manually.")
+                        # Optionally show the exception text for debugging
+                        st.exception(e)
+                        # Stop processing this user input
+                        return
+
+                    # Attempt an automatic rebuild and retry once
+                    attempted_rebuild = True
+                    st.warning("Detected retrieval backend issue — attempting to rebuild the vectorstore and retry...")
+                    try:
+                        run_ingest_cli(data_dir=self.data_dir, persist_dir=self.persist_dir)
+                    except Exception as rebuild_err:
+                        st.error("Automatic rebuild failed; please rebuild manually from the sidebar or CLI.")
+                        st.exception(rebuild_err)
+                        return
+                    # Clear cached retriever and chain and reload
+                    try:
+                        build_or_load_retriever_cached.clear()
+                        get_chain_cached.clear()
+                        self.chain = get_chain_cached(
+                            model_name=self.model_name,
+                            top_k=self.top_k,
+                            retrieval_mode=self.retrieval_mode,
+                            data_dir=self.data_dir,
+                            persist_dir=self.persist_dir,
+                        )
+                    except Exception as reload_err:
+                        st.error("Failed to reload the QA chain after rebuilding the vectorstore.")
+                        st.exception(reload_err)
+                        return
+                    # loop will retry once
 
             answer = (
                     result.get("answer")