voice-vibecoder 2.17.6__tar.gz → 2.18.0__tar.gz

{voice_vibecoder-2.17.6 → voice_vibecoder-2.18.0}/CHANGELOG.md

@@ -2,6 +2,14 @@
 
 <!-- version list -->
 
+## v2.18.0 (2026-02-23)
+
+### Features
+
+- **server**: Text-only mode bypasses voice provider
+  ([`1b62fcf`](https://github.com/snokam/voice-vibecoder/commit/1b62fcfffa15ce44a911781761610c533a69bac8))
+
+
 ## v2.17.6 (2026-02-23)
 
 ### Bug Fixes

{voice_vibecoder-2.17.6 → voice_vibecoder-2.18.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: voice-vibecoder
-Version: 2.17.6
+Version: 2.18.0
 Summary: Olaf The Vibecoder — a voice-controlled coding assistant using OpenAI Realtime API + Claude Code
 Project-URL: Homepage, https://github.com/snokam/voice-vibecoder
 Project-URL: Repository, https://github.com/snokam/voice-vibecoder

{voice_vibecoder-2.17.6 → voice_vibecoder-2.18.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "voice-vibecoder"
-version = "2.17.6"
+version = "2.18.0"
 description = "Olaf The Vibecoder — a voice-controlled coding assistant using OpenAI Realtime API + Claude Code"
 readme = "README.md"
 requires-python = ">=3.10"

{voice_vibecoder-2.17.6 → voice_vibecoder-2.18.0}/src/voice_vibecoder/__init__.py

@@ -3,7 +3,7 @@
 A voice-controlled coding assistant using OpenAI Realtime API + Claude Code.
 """
 
-__version__ = "2.17.6"
+__version__ = "2.18.0"
 
 from voice_vibecoder.app_config import VoiceConfig
 from voice_vibecoder.ui.screen import VoiceCodingScreen

{voice_vibecoder-2.17.6 → voice_vibecoder-2.18.0}/src/voice_vibecoder/server.py

@@ -132,18 +132,16 @@ class WebSocketBridge:
             on_ui_command=self._on_ui_command,
         )
 
+        # Set event loop so _send works even before voice provider connects
+        self._session._loop = asyncio.get_event_loop()
+
         # Load saved instances — panels appear dynamically as sessions/instances are created
         restored = self._session.registry.load_state()
 
-        await self._session.connect()
-        await self._session.configure_session()
-        # Don't sync state here — the frontend sends sync_state after
-        # connecting (and after auth), which triggers _sync_state_to_frontend.
-        # Sending it here too would cause duplicate output lines.
+        # Don't connect voice provider yet — wait for sync_state with audio: true.
+        # Text-only clients skip the voice provider entirely.
         self._send({"type": "status", "status": "ready"})
 
-        self._messages_task = asyncio.create_task(self._session.handle_messages())
-
         try:
             await self._receive_loop()
         finally:
@@ -156,6 +154,14 @@ class WebSocketBridge:
             if self._session.connected:
                 await self._session.disconnect()
 
+    async def _connect_voice_provider(self) -> None:
+        """Lazily connect the voice provider when audio is needed."""
+        if self._session.connected:
+            return
+        await self._session.connect()
+        await self._session.configure_session()
+        self._messages_task = asyncio.create_task(self._session.handle_messages())
+
     async def _receive_loop(self) -> None:
         try:
             async for raw in self._ws:
@@ -167,32 +173,48 @@ class WebSocketBridge:
                 msg_type = msg.get("type")
                 if msg_type == "audio":
                     data = msg.get("data", "")
-                    if data and self._session:
+                    if data and self._session and self._session.connected:
                         await self._session.send_audio(data)
                 elif msg_type == "audio.commit":
-                    if self._session:
+                    if self._session and self._session.connected:
                         await self._session.commit_audio()
                 elif msg_type == "text":
-                    # Text-only mode: inject user message directly
                     text = msg.get("text", "") or msg.get("message", "")
                     if text and self._session:
-                        # Transcript sent by session.inject_user_message
-                        await self._session.inject_user_message(text)
+                        if self._session.connected:
+                            # Voice mode: route through voice provider
+                            await self._session.inject_user_message(text)
+                        else:
+                            # Text-only mode: route directly to agent
+                            await self._handle_text_only(text)
                 elif msg_type == "update_settings":
                     await self._handle_settings_update(msg.get("settings", {}))
                 elif msg_type == "sync_state":
-                    # Frontend requests full state sync from backend.
-                    # Opt-in to ui_command messages (file_tree, show_diff, etc.)
+                    # Opt-in flags
                     if msg.get("ui_commands"):
                         self._ui_commands_enabled = True
                     if msg.get("audio"):
                         self._audio_enabled = True
+                        # Connect voice provider now (lazy)
+                        await self._connect_voice_provider()
                     self._send({"type": "status", "status": "syncing"})
                     await self._sync_state_to_frontend()
                     self._send({"type": "status", "status": "ready"})
         except websockets.ConnectionClosed:
             pass
 
+    async def _handle_text_only(self, text: str) -> None:
+        """Route text directly to an agent session, bypassing the voice provider.
+
+        Used for text-only clients that don't need audio. The agent streams
+        output via on_agent_output and sends the final result as a transcript.
+        """
+        from voice_vibecoder.tools.agent_task import _handle_send_to_session
+
+        self._session.text_only = True
+        self._send({"type": "transcript", "role": "user", "text": text})
+        _handle_send_to_session(text)
+
     async def _sync_state_to_frontend(self) -> None:
         """Send full backend state to frontend - all instances with their conversation history."""
         if not self._session:

{voice_vibecoder-2.17.6 → voice_vibecoder-2.18.0}/src/voice_vibecoder/session.py

@@ -85,6 +85,7 @@ class RealtimeSession:
         self._on_ui_command = on_ui_command
         self._running = False
         self._assistant_transcript = ""
+        self.text_only = False
         self._response_pending = False
         self._fallback_task: asyncio.Task | None = None
         self._loop: asyncio.AbstractEventLoop | None = None
@@ -358,6 +359,11 @@ class RealtimeSession:
             # Show only the first line as summary — full result was already streamed
             first_line = result.strip().split("\n", 1)[0][:120]
             self._registry.on_output(instance_id, f"{prefix} {first_line}")
+        if self.text_only:
+            # Text-only mode: send result as transcript directly (no voice)
+            if self._on_transcript:
+                self._on_transcript("assistant", result)
+            return
         if self._loop and self._running:
             inst = self._registry.get_by_id(instance_id)
             display_key = inst.display_key if inst else "unknown"