feat read files

gui.py

@@ -439,7 +439,7 @@ class ModelLoaderThread(QThread):
 
 
 class AskThread(QThread):
-    tool_call = pyqtSignal(str)
+    tool_call = pyqtSignal(str, str)
     usage = pyqtSignal(dict)
     token = pyqtSignal(str)
     thinking_token = pyqtSignal(str)
@@ -479,6 +479,11 @@ class ChatWindow(QMainWindow):
             self._sessions[0] if self._sessions else chatbot.new_session()
         )
         self._history: list[tuple[str, str]] = list(self._current_session.history)
+        # The loaded model has a single KV cache. Each chat keeps its own
+        # snapshot of that cache so switching chats restores that chat's
+        # context instead of continuing whichever chat ran last.
+        self._context_states: dict[str, object] = {}
+        self._context_owner: str | None = None
         self._pending_question = ""
         self._ask_thread: AskThread | None = None
         # Position in chat_view's document where the in-progress streamed
@@ -718,6 +723,7 @@ class ChatWindow(QMainWindow):
             return  # already on an empty, unsaved chat
         self._current_session = chatbot.new_session()
         self._history = []
+        self._activate_session_context(self._current_session)
         self.chat_view.clear()
         self._append_system("Dev assist chatbot ready. Ask a development question.")
         self.session_list.clearSelection()
@@ -734,6 +740,7 @@ class ChatWindow(QMainWindow):
             return
         self._current_session = session
         self._history = list(session.history)
+        self._activate_session_context(session)
         self._render_session_history(session)
         self.statusBar().showMessage("Ready.")
 
@@ -759,6 +766,9 @@ class ChatWindow(QMainWindow):
         if reply != QMessageBox.StandardButton.Yes:
             return
         chatbot.delete_session(session_id)
+        self._context_states.pop(session_id, None)
+        if self._context_owner == session_id:
+            self._context_owner = None
         self._sessions = [s for s in self._sessions if s.id != session_id]
         if self._current_session.id == session_id:
             if self._sessions:
@@ -766,6 +776,7 @@ class ChatWindow(QMainWindow):
             else:
                 self._current_session = chatbot.new_session()
             self._history = list(self._current_session.history)
+            self._activate_session_context(self._current_session)
             self.chat_view.clear()
             if self._current_session.history:
                 self._render_session_history(self._current_session)
@@ -776,7 +787,7 @@ class ChatWindow(QMainWindow):
     def _on_model_loaded(self, llm) -> None:
         self._llm = llm
         self._n_ctx = llm.n_ctx()
-        self.context_status_label.setText(f"Context: 0 / {self._n_ctx}")
+        self._activate_session_context(self._current_session)
         self.statusBar().showMessage("Ready.")
         self._append_system("Dev assist chatbot ready. Ask a development question.")
         self._set_input_enabled(True)
@@ -799,6 +810,7 @@ class ChatWindow(QMainWindow):
         self._answer_started = False
         self._stream_thinking_id = None
 
+        self._activate_session_context(self._current_session)
         messages = chatbot.build_messages(self._history, question)
         self._pending_question = question
 
@@ -936,16 +948,47 @@ class ChatWindow(QMainWindow):
             return
         self._render_thinking(block_id, expanded=not block["expanded"])
 
-    def _on_tool_call(self, query: str) -> None:
+    def _on_tool_call(self, tool_name: str, target: str) -> None:
         self._collapse_open_thinking()
         self._discard_stream_card()
-        self.statusBar().showMessage(f"Searching: {query}")
-        self._append_system(f"Searching the web for: {query}")
+        if tool_name == "read_file":
+            self.statusBar().showMessage(f"Reading file: {target}")
+            self._append_system(f"Reading file: {target}")
+        else:
+            self.statusBar().showMessage(f"Searching: {target}")
+            self._append_system(f"Searching the web for: {target}")
+
+    def _show_context_usage(self, total: int | None) -> None:
+        if total is None or not self._n_ctx:
+            shown = "--" if total is None else total
+            limit = self._n_ctx or "--"
+            self.context_status_label.setText(f"Context: {shown} / {limit}")
+            return
+        pct = 100 * total / self._n_ctx
+        self.context_status_label.setText(f"Context: {total} / {self._n_ctx} ({pct:.0f}%)")
+
+    def _activate_session_context(self, session: "chatbot.ChatSession") -> None:
+        """Make the loaded model's KV cache match this chat.
+
+        Leaving a chat stores its cache; entering one restores it, or clears
+        the cache when that chat has no snapshot yet (first visit this run).
+        """
+        self._show_context_usage(session.context_tokens)
+        if self._llm is None or self._context_owner == session.id:
+            return
+        if self._context_owner is not None:
+            self._context_states[self._context_owner] = self._llm.save_state()
+        state = self._context_states.get(session.id)
+        if state is None:
+            self._llm.reset()
+        else:
+            self._llm.load_state(state)
+        self._context_owner = session.id
 
     def _on_usage(self, usage: dict) -> None:
         total = usage["total_tokens"]
-        pct = 100 * total / self._n_ctx if self._n_ctx else 0
-        self.context_status_label.setText(f"Context: {total} / {self._n_ctx} ({pct:.0f}%)")
+        self._current_session.context_tokens = total
+        self._show_context_usage(total)
 
     def _on_answer(self, answer: str) -> None:
         # Normally already collapsed by the first answer token (_on_token),