Add cloud-only sidecar variant (~50MB vs 500MB-2GB)

Lightweight Deepgram-only sidecar that excludes PyTorch, faster-whisper, RealtimeSTT, and CUDA. Only includes audio capture + WebSocket streaming to Deepgram. Requires a Deepgram API key (BYOK or managed mode). Changes: - client/models.py: Extracted TranscriptionResult into standalone module so deepgram_transcription.py doesn't transitively import torch - backend/app_controller.py: Made RealtimeTranscriptionEngine and DeviceManager imports lazy (only loaded when remote.mode == "local") - local-transcription-cloud.spec: PyInstaller spec excluding all ML deps - SidecarSetup.svelte: Added "Cloud Only (Deepgram)" variant option - build-sidecar-cloud.yml: CI workflow building cloud sidecar for all 3 OS - sidecar-release.yml: Dispatches cloud build alongside CPU/CUDA builds Sidecar download options are now: - Standard (CPU): ~500 MB - local Whisper on any computer - GPU Accelerated (CUDA): ~2 GB - local Whisper with NVIDIA GPU - Cloud Only (Deepgram): ~50 MB - requires API key, no local models Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-07 16:57:43 -07:00
parent bb039399fc
commit 3d3d7ec3c5
10 changed files with 469 additions and 42 deletions
--- a/.gitea/workflows/build-sidecar-cloud.yml
+++ b/.gitea/workflows/build-sidecar-cloud.yml
@@ -0,0 +1,227 @@
+name: Build Sidecar (Cloud)
+
+on:
+  workflow_dispatch:
+    inputs:
+      tag:
+        description: 'Sidecar release tag to build (e.g. sidecar-v1.0.5)'
+        required: true
+
+jobs:
+  build-cloud-linux:
+    name: Build Cloud Sidecar (Linux)
+    runs-on: ubuntu-latest
+    env:
+      PYTHON_VERSION: "3.11"
+      RELEASE_TAG: ${{ inputs.tag }}
+    steps:
+      - name: Show tag
+        run: echo "Building cloud sidecar for tag ${RELEASE_TAG}"
+
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.tag }}
+
+      - name: Install uv
+        run: |
+          curl -LsSf https://astral.sh/uv/install.sh | sh
+          echo "$HOME/.local/bin" >> $GITHUB_PATH
+
+      - name: Set up Python
+        run: uv python install ${{ env.PYTHON_VERSION }}
+
+      - name: Install system dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y portaudio19-dev
+
+      - name: Build cloud sidecar
+        env:
+          UV_NO_SOURCES: "1"
+        run: |
+          uv venv
+          uv pip install pyinstaller numpy sounddevice fastapi uvicorn websockets pydantic requests pyyaml packaging
+          .venv/bin/pyinstaller local-transcription-cloud.spec
+
+      - name: Package
+        run: |
+          cd dist/local-transcription-backend && zip -r ../../sidecar-linux-x86_64-cloud.zip .
+
+      - name: Upload to release
+        env:
+          BUILD_TOKEN: ${{ secrets.BUILD_TOKEN }}
+        run: |
+          sudo apt-get install -y jq
+          REPO_API="${GITHUB_SERVER_URL}/api/v1/repos/${GITHUB_REPOSITORY}"
+          TAG="${RELEASE_TAG}"
+
+          for i in $(seq 1 30); do
+            RELEASE_ID=$(curl -s -H "Authorization: token ${BUILD_TOKEN}" \
+              "${REPO_API}/releases/tags/${TAG}" | jq -r '.id // empty')
+            if [ -n "${RELEASE_ID}" ] && [ "${RELEASE_ID}" != "null" ]; then
+              echo "Found release ${TAG} (ID: ${RELEASE_ID})"
+              break
+            fi
+            echo "Attempt ${i}/30: waiting for release..."
+            sleep 10
+          done
+
+          if [ -z "${RELEASE_ID}" ] || [ "${RELEASE_ID}" = "null" ]; then
+            echo "ERROR: Release not found"; exit 1
+          fi
+
+          for file in sidecar-*-cloud.zip; do
+            filename=$(basename "$file")
+            ASSET_ID=$(curl -s -H "Authorization: token ${BUILD_TOKEN}" \
+              "${REPO_API}/releases/${RELEASE_ID}/assets" | jq -r ".[] | select(.name == \"${filename}\") | .id // empty")
+            [ -n "${ASSET_ID}" ] && curl -s -X DELETE -H "Authorization: token ${BUILD_TOKEN}" "${REPO_API}/releases/${RELEASE_ID}/assets/${ASSET_ID}"
+            curl -s -o /dev/null -w "Upload ${filename}: HTTP %{http_code}\n" -X POST \
+              -H "Authorization: token ${BUILD_TOKEN}" -H "Content-Type: application/octet-stream" \
+              -T "$file" "${REPO_API}/releases/${RELEASE_ID}/assets?name=${filename}"
+          done
+
+  build-cloud-windows:
+    name: Build Cloud Sidecar (Windows)
+    runs-on: windows-latest
+    env:
+      PYTHON_VERSION: "3.11"
+      RELEASE_TAG: ${{ inputs.tag }}
+    steps:
+      - name: Show tag
+        shell: powershell
+        run: Write-Host "Building cloud sidecar for tag $env:RELEASE_TAG"
+
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.tag }}
+
+      - name: Install uv
+        shell: powershell
+        run: |
+          if (Get-Command uv -ErrorAction SilentlyContinue) {
+            Write-Host "uv already installed"
+          } else {
+            irm https://astral.sh/uv/install.ps1 | iex
+            $uvPaths = @("$env:USERPROFILE\.local\bin", "$env:USERPROFILE\.cargo\bin", "$env:LOCALAPPDATA\uv\bin")
+            foreach ($p in $uvPaths) { if (Test-Path $p) { echo $p | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append } }
+          }
+
+      - name: Set up Python
+        shell: powershell
+        run: uv python install ${{ env.PYTHON_VERSION }}
+
+      - name: Build cloud sidecar
+        shell: powershell
+        env:
+          UV_NO_SOURCES: "1"
+        run: |
+          uv venv
+          uv pip install pyinstaller numpy sounddevice fastapi uvicorn websockets pydantic requests pyyaml packaging
+          .venv\Scripts\pyinstaller.exe local-transcription-cloud.spec
+
+      - name: Package
+        shell: powershell
+        run: |
+          if (-not (Get-Command 7z -ErrorAction SilentlyContinue)) { choco install 7zip -y }
+          7z a -tzip -mx=5 sidecar-windows-x86_64-cloud.zip .\dist\local-transcription-backend\*
+
+      - name: Upload to release
+        shell: powershell
+        env:
+          BUILD_TOKEN: ${{ secrets.BUILD_TOKEN }}
+        run: |
+          $REPO_API = "${{ github.server_url }}/api/v1/repos/${{ github.repository }}"
+          $Headers = @{ "Authorization" = "token $env:BUILD_TOKEN" }
+          $TAG = $env:RELEASE_TAG
+
+          $RELEASE_ID = $null
+          for ($i = 1; $i -le 30; $i++) {
+            try {
+              $release = Invoke-RestMethod -Uri "$REPO_API/releases/tags/$TAG" -Headers $Headers -ErrorAction Stop
+              $RELEASE_ID = $release.id
+              if ($RELEASE_ID) { Write-Host "Found release $TAG (ID: $RELEASE_ID)"; break }
+            } catch {}
+            Write-Host "Attempt ${i}/30: waiting..."; Start-Sleep -Seconds 10
+          }
+          if (-not $RELEASE_ID) { Write-Host "ERROR: Release not found"; exit 1 }
+
+          Get-ChildItem -Path . -Filter "sidecar-*-cloud.zip" | ForEach-Object {
+            $fn = $_.Name; $enc = [System.Uri]::EscapeDataString($fn)
+            try {
+              $assets = Invoke-RestMethod -Uri "$REPO_API/releases/$RELEASE_ID/assets" -Headers $Headers
+              $existing = $assets | Where-Object { $_.name -eq $fn }
+              if ($existing) { Invoke-RestMethod -Uri "$REPO_API/releases/$RELEASE_ID/assets/$($existing.id)" -Method Delete -Headers $Headers }
+            } catch {}
+            curl.exe --fail -s -X POST -H "Authorization: token $env:BUILD_TOKEN" -H "Content-Type: application/octet-stream" -T "$($_.FullName)" "$REPO_API/releases/$RELEASE_ID/assets?name=$enc"
+            Write-Host "Uploaded $fn"
+          }
+
+  build-cloud-macos:
+    name: Build Cloud Sidecar (macOS)
+    runs-on: macos-latest
+    env:
+      PYTHON_VERSION: "3.11"
+      RELEASE_TAG: ${{ inputs.tag }}
+    steps:
+      - name: Show tag
+        run: echo "Building cloud sidecar for tag ${RELEASE_TAG}"
+
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.tag }}
+
+      - name: Install uv
+        run: |
+          curl -LsSf https://astral.sh/uv/install.sh | sh
+          echo "$HOME/.local/bin" >> $GITHUB_PATH
+
+      - name: Set up Python
+        run: uv python install ${{ env.PYTHON_VERSION }}
+
+      - name: Install system dependencies
+        run: brew install portaudio
+
+      - name: Build cloud sidecar
+        env:
+          UV_NO_SOURCES: "1"
+        run: |
+          uv venv
+          uv pip install pyinstaller numpy sounddevice fastapi uvicorn websockets pydantic requests pyyaml packaging
+          .venv/bin/pyinstaller local-transcription-cloud.spec
+
+      - name: Package
+        run: |
+          cd dist/local-transcription-backend && zip -r ../../sidecar-macos-aarch64-cloud.zip .
+
+      - name: Upload to release
+        env:
+          BUILD_TOKEN: ${{ secrets.BUILD_TOKEN }}
+        run: |
+          which jq || brew install jq
+          REPO_API="${GITHUB_SERVER_URL}/api/v1/repos/${GITHUB_REPOSITORY}"
+          TAG="${RELEASE_TAG}"
+
+          for i in $(seq 1 30); do
+            RELEASE_ID=$(curl -s -H "Authorization: token ${BUILD_TOKEN}" \
+              "${REPO_API}/releases/tags/${TAG}" | jq -r '.id // empty')
+            if [ -n "${RELEASE_ID}" ] && [ "${RELEASE_ID}" != "null" ]; then
+              echo "Found release ${TAG} (ID: ${RELEASE_ID})"
+              break
+            fi
+            echo "Attempt ${i}/30: waiting for release..."
+            sleep 10
+          done
+
+          if [ -z "${RELEASE_ID}" ] || [ "${RELEASE_ID}" = "null" ]; then
+            echo "ERROR: Release not found"; exit 1
+          fi
+
+          for file in sidecar-*-cloud.zip; do
+            filename=$(basename "$file")
+            ASSET_ID=$(curl -s -H "Authorization: token ${BUILD_TOKEN}" \
+              "${REPO_API}/releases/${RELEASE_ID}/assets" | jq -r ".[] | select(.name == \"${filename}\") | .id // empty")
+            [ -n "${ASSET_ID}" ] && curl -s -X DELETE -H "Authorization: token ${BUILD_TOKEN}" "${REPO_API}/releases/${RELEASE_ID}/assets/${ASSET_ID}"
+            curl -s -o /dev/null -w "Upload ${filename}: HTTP %{http_code}\n" -X POST \
+              -H "Authorization: token ${BUILD_TOKEN}" -H "Content-Type: application/octet-stream" \
+              -T "$file" "${REPO_API}/releases/${RELEASE_ID}/assets?name=${filename}"
+          done
--- a/.gitea/workflows/sidecar-release.yml
+++ b/.gitea/workflows/sidecar-release.yml
@@ -118,7 +118,7 @@ jobs:
          REPO_API="${GITHUB_SERVER_URL}/api/v1/repos/${GITHUB_REPOSITORY}"
          TAG="${{ steps.bump.outputs.tag }}"

-          for workflow in build-sidecar-linux.yml build-sidecar-windows.yml build-sidecar-macos.yml; do
+          for workflow in build-sidecar-linux.yml build-sidecar-windows.yml build-sidecar-macos.yml build-sidecar-cloud.yml; do
            echo "Dispatching ${workflow} for ${TAG}..."
            HTTP_CODE=$(curl -s -o /dev/null -w "%{http_code}" -X POST \
              -H "Authorization: token ${BUILD_TOKEN}" \
--- a/backend/app_controller.py
+++ b/backend/app_controller.py
@@ -18,13 +18,18 @@ import sys
 sys.path.insert(0, str(Path(__file__).resolve().parent.parent))

 from client.config import Config
-from client.device_utils import DeviceManager
-from client.transcription_engine_realtime import RealtimeTranscriptionEngine, TranscriptionResult
+from client.models import TranscriptionResult
 from client.deepgram_transcription import DeepgramTranscriptionEngine
 from client.server_sync import ServerSyncClient
 from server.web_display import TranscriptionWebServer
 from version import __version__

+# Heavy imports (torch, RealtimeSTT, faster-whisper) are deferred so
+# the cloud-only sidecar build can exclude them entirely.
+# Imported lazily in _initialize_engine() when remote.mode == "local".
+RealtimeTranscriptionEngine = None
+DeviceManager = None
+

 class AppState:
    """Enum-like class for application states."""
@@ -89,7 +94,18 @@ class AppController:

    def __init__(self, config: Optional[Config] = None):
        self.config = config or Config()
-        self.device_manager = DeviceManager()
+
+        # DeviceManager is only needed for local Whisper mode.
+        # Lazy-import to keep the cloud-only sidecar lightweight.
+        global DeviceManager
+        if DeviceManager is None:
+            try:
+                from client.device_utils import DeviceManager as _DM
+                DeviceManager = _DM
+            except ImportError:
+                DeviceManager = None
+
+        self.device_manager = DeviceManager() if DeviceManager else None

        # State
        self._state = AppState.INITIALIZING
@@ -243,15 +259,12 @@ class AppController:

    def _initialize_engine(self):
        """Initialize the transcription engine in a background thread."""
-        device_config = self.config.get('transcription.device', 'auto')
-        self.device_manager.set_device(device_config)
-
        audio_device_str = self.config.get('audio.input_device', 'default')
        audio_device = None if audio_device_str == 'default' else int(audio_device_str)

        model = self.config.get('transcription.model', 'base.en')
        language = self.config.get('transcription.language', 'en')
-        device = self.device_manager.get_device_for_whisper()
+        device_config = self.config.get('transcription.device', 'auto')
        compute_type = self.config.get('transcription.compute_type', 'default')

        self.current_model_size = model
@@ -284,6 +297,18 @@ class AppController:
            self.transcription_engine.set_error_callback(self._on_remote_error)
            self.transcription_engine.set_credits_low_callback(self._on_credits_low)
        else:
+            # Lazy-import heavy local transcription dependencies
+            global RealtimeTranscriptionEngine
+            if RealtimeTranscriptionEngine is None:
+                from client.transcription_engine_realtime import RealtimeTranscriptionEngine as _RTE
+                RealtimeTranscriptionEngine = _RTE
+
+            if self.device_manager:
+                self.device_manager.set_device(device_config)
+                device = self.device_manager.get_device_for_whisper()
+            else:
+                device = "cpu"
+
            self.transcription_engine = RealtimeTranscriptionEngine(
                model=model,
                device=device,
@@ -602,7 +627,7 @@ class AppController:
        host = self.config.get('web_server.host', '127.0.0.1')
        port = self.actual_web_port or self.config.get('web_server.port', 8080)

-        device_info = self.device_manager.get_device_info()
+        device_info = self.device_manager.get_device_info() if self.device_manager else []

        remote_mode = self.config.get('remote.mode', 'local')
        if remote_mode in ('managed', 'byok') and self.transcription_engine:
@@ -646,10 +671,13 @@ class AppController:

    def get_compute_devices(self) -> list[dict]:
        """List available compute devices."""
-        device_info = self.device_manager.get_device_info()
        devices = [{"id": "auto", "name": "Auto-detect"}]
+        if self.device_manager:
+            device_info = self.device_manager.get_device_info()
            for dev_id, dev_name in device_info:
                devices.append({"id": dev_id, "name": dev_name})
+        else:
+            devices.append({"id": "cloud", "name": "Cloud (Deepgram)"})
        return devices

    # ── Update Checking ────────────────────────────────────────────
--- a/backend/tests/test_api_server.py
+++ b/backend/tests/test_api_server.py
@@ -79,7 +79,7 @@ async def test_start_when_not_ready(api_client, controller):

@pytest.mark.asyncio
 async def test_clear(api_client, controller):
-    from client.transcription_engine_realtime import TranscriptionResult
+    from client.models import TranscriptionResult
    from datetime import datetime

    controller.transcriptions = [
--- a/backend/tests/test_app_controller.py
+++ b/backend/tests/test_app_controller.py
@@ -72,7 +72,7 @@ def test_double_start_rejected(controller):

 def test_clear_transcriptions(controller):
    """clear_transcriptions should empty the list and return the count."""
-    from client.transcription_engine_realtime import TranscriptionResult
+    from client.models import TranscriptionResult

    controller.transcriptions = [
        TranscriptionResult(text="Hello", is_final=True, timestamp=datetime.now(), user_name="Alice"),
@@ -85,7 +85,7 @@ def test_clear_transcriptions(controller):

 def test_get_transcriptions_text_with_timestamps(controller):
    """get_transcriptions_text should include [HH:MM:SS] prefixes when requested."""
-    from client.transcription_engine_realtime import TranscriptionResult
+    from client.models import TranscriptionResult

    ts = datetime(2025, 1, 15, 10, 30, 45)
    controller.transcriptions = [
@@ -141,7 +141,7 @@ def test_apply_settings_no_reload_when_same(controller):

 def test_on_final_transcription_callback_fires(controller):
    """_on_final_transcription should append and invoke on_transcription callback."""
-    from client.transcription_engine_realtime import TranscriptionResult
+    from client.models import TranscriptionResult

    received = []
    controller.on_transcription = lambda data: received.append(data)
@@ -166,7 +166,7 @@ def test_on_final_transcription_callback_fires(controller):

 def test_on_final_transcription_ignored_when_not_transcribing(controller):
    """If the controller is not in transcribing state the callback should be a no-op."""
-    from client.transcription_engine_realtime import TranscriptionResult
+    from client.models import TranscriptionResult

    controller.is_transcribing = False

--- a/client/deepgram_transcription.py
+++ b/client/deepgram_transcription.py
@@ -17,7 +17,7 @@ from datetime import datetime
 from queue import Queue, Empty
 from typing import Optional, Callable

-from client.transcription_engine_realtime import TranscriptionResult
+from client.models import TranscriptionResult

 logger = logging.getLogger(__name__)

--- a/client/models.py
+++ b/client/models.py
@@ -0,0 +1,29 @@
+"""Shared data models used across transcription engines."""
+
+from datetime import datetime
+
+
+class TranscriptionResult:
+    """Represents a transcription result."""
+
+    def __init__(self, text: str, is_final: bool, timestamp: datetime, user_name: str = ""):
+        """
+        Initialize transcription result.
+
+        Args:
+            text: Transcribed text
+            is_final: Whether this is a final transcription or realtime preview
+            timestamp: Timestamp of transcription
+            user_name: Name of the user/speaker
+        """
+        self.text = text.strip()
+        self.is_final = is_final
+        self.timestamp = timestamp
+        self.user_name = user_name
+
+    def __repr__(self) -> str:
+        time_str = self.timestamp.strftime("%H:%M:%S")
+        prefix = "[FINAL]" if self.is_final else "[PREVIEW]"
+        if self.user_name and self.user_name.strip():
+            return f"{prefix} [{time_str}] {self.user_name}: {self.text}"
+        return f"{prefix} [{time_str}] {self.text}"
--- a/client/transcription_engine_realtime.py
+++ b/client/transcription_engine_realtime.py
@@ -8,30 +8,8 @@ from threading import Lock
 import logging


-class TranscriptionResult:
-    """Represents a transcription result."""
-
-    def __init__(self, text: str, is_final: bool, timestamp: datetime, user_name: str = ""):
-        """
-        Initialize transcription result.
-
-        Args:
-            text: Transcribed text
-            is_final: Whether this is a final transcription or realtime preview
-            timestamp: Timestamp of transcription
-            user_name: Name of the user/speaker
-        """
-        self.text = text.strip()
-        self.is_final = is_final
-        self.timestamp = timestamp
-        self.user_name = user_name
-
-    def __repr__(self) -> str:
-        time_str = self.timestamp.strftime("%H:%M:%S")
-        prefix = "[FINAL]" if self.is_final else "[PREVIEW]"
-        if self.user_name and self.user_name.strip():
-            return f"{prefix} [{time_str}] {self.user_name}: {self.text}"
-        return f"{prefix} [{time_str}] {self.text}"
+# Re-export TranscriptionResult from the shared models module for backward compatibility
+from client.models import TranscriptionResult  # noqa: F401

    def to_dict(self) -> dict:
        """Convert to dictionary."""
--- a/local-transcription-cloud.spec
+++ b/local-transcription-cloud.spec
@@ -0,0 +1,152 @@
+# -*- mode: python ; coding: utf-8 -*-
+"""PyInstaller spec file for cloud-only Local Transcription backend.
+
+This builds a lightweight sidecar (~50MB) that only supports Deepgram
+cloud transcription (managed + BYOK). No local Whisper models, no
+PyTorch, no CUDA -- just audio capture and WebSocket streaming.
+"""
+
+import sys
+import os
+
+block_cipher = None
+is_windows = sys.platform == 'win32'
+
+from PyInstaller.utils.hooks import collect_submodules, collect_data_files
+
+# Data files
+datas = [
+    ('config/default_config.yaml', 'config'),
+]
+
+# Hidden imports -- only lightweight deps needed for Deepgram streaming
+hiddenimports = [
+    'sounddevice',
+    'numpy',
+    # FastAPI and dependencies
+    'fastapi',
+    'fastapi.routing',
+    'fastapi.responses',
+    'starlette',
+    'starlette.applications',
+    'starlette.routing',
+    'starlette.responses',
+    'starlette.websockets',
+    'starlette.middleware',
+    'starlette.middleware.cors',
+    'pydantic',
+    'pydantic.fields',
+    'pydantic.main',
+    'anyio',
+    'anyio._backends',
+    'anyio._backends._asyncio',
+    'sniffio',
+    # Uvicorn
+    'uvicorn',
+    'uvicorn.logging',
+    'uvicorn.loops',
+    'uvicorn.loops.auto',
+    'uvicorn.protocols',
+    'uvicorn.protocols.http',
+    'uvicorn.protocols.http.auto',
+    'uvicorn.protocols.http.h11_impl',
+    'uvicorn.protocols.websockets',
+    'uvicorn.protocols.websockets.auto',
+    'uvicorn.protocols.websockets.wsproto_impl',
+    'uvicorn.lifespan',
+    'uvicorn.lifespan.on',
+    'h11',
+    'websockets',
+    'websockets.legacy',
+    'websockets.legacy.server',
+    # HTTP client
+    'requests',
+    'urllib3',
+    'certifi',
+    'charset_normalizer',
+]
+
+# Collect submodules for key packages
+print("Collecting submodules for cloud backend packages...")
+for package in ['fastapi', 'starlette', 'pydantic', 'pydantic_core', 'anyio', 'uvicorn', 'websockets', 'h11']:
+    try:
+        submodules = collect_submodules(package)
+        hiddenimports += submodules
+        print(f"  + Collected {len(submodules)} submodules from {package}")
+    except Exception as e:
+        print(f"  - Warning: Could not collect {package}: {e}")
+
+# Collect data files
+for package in ['fastapi', 'starlette', 'pydantic', 'uvicorn']:
+    try:
+        data_files = collect_data_files(package)
+        if data_files:
+            datas += data_files
+    except Exception:
+        pass
+
+# Pydantic critical deps
+hiddenimports += [
+    'colorsys', 'decimal', 'json', 'ipaddress', 'pathlib', 'uuid',
+    'email.message', 'typing_extensions',
+]
+
+a = Analysis(
+    ['backend/main_headless.py'],
+    pathex=[],
+    binaries=[],
+    datas=datas,
+    hiddenimports=hiddenimports,
+    hookspath=['hooks'],
+    hooksconfig={},
+    runtime_hooks=[],
+    excludes=[
+        # Exclude all heavy ML/local transcription deps
+        'torch', 'torchaudio', 'torchvision',
+        'faster_whisper', 'ctranslate2',
+        'RealtimeSTT', 'webrtcvad', 'webrtcvad_wheels',
+        'silero_vad', 'onnxruntime',
+        'openwakeword', 'pvporcupine', 'pyaudio',
+        'noisereduce', 'scipy',
+        # Exclude GUI frameworks
+        'PySide6', 'PyQt5', 'PyQt6', 'tkinter',
+        # Exclude other unnecessary heavy packages
+        'matplotlib', 'PIL', 'cv2',
+    ],
+    win_no_prefer_redirects=False,
+    win_private_assemblies=False,
+    cipher=block_cipher,
+    noarchive=False,
+)
+
+pyz = PYZ(a.pure, a.zipped_data, cipher=block_cipher)
+
+exe = EXE(
+    pyz,
+    a.scripts,
+    [],
+    exclude_binaries=True,
+    name='local-transcription-backend',
+    debug=False,
+    bootloader_ignore_signals=False,
+    strip=False,
+    upx=True,
+    console=True,
+    disable_windowed_traceback=False,
+    argv_emulation=False,
+    target_arch=None,
+    codesign_identity=None,
+    entitlements_file=None,
+    icon='LocalTranscription.ico' if is_windows else None,
+)
+
+coll = COLLECT(
+    exe,
+    a.binaries,
+    a.zipfiles,
+    a.datas,
+    strip=False,
+    upx=True,
+    upx_exclude=[],
+    name='local-transcription-backend',
+)
--- a/src/lib/components/SidecarSetup.svelte
+++ b/src/lib/components/SidecarSetup.svelte
@@ -114,6 +114,19 @@
            <span class="variant-desc">Faster transcription with NVIDIA GPU (~2 GB download)</span>
          </div>
        </label>
+
+        <label class="variant-option" class:selected={variant === "cloud"}>
+          <input
+            type="radio"
+            name="variant"
+            value="cloud"
+            bind:group={variant}
+          />
+          <div class="variant-info">
+            <span class="variant-name">Cloud Only (Deepgram)</span>
+            <span class="variant-desc">Lightweight, requires Deepgram API key (~50 MB download)</span>
+          </div>
+        </label>
      </div>

      <button class="download-btn" onclick={startDownload}>