Merge branch 'experimental'

README.md

@@ -128,6 +128,22 @@ source venv_all/bin/activate
 python coderai          # starts on http://127.0.0.1:8776
 ```
 
+macOS:
+
+```bash
+./osxbuild.sh all
+source venv_osx_all/bin/activate
+python coderai
+```
+
+Windows PowerShell:
+
+```powershell
+.\build.ps1 -Backend all
+.\venv_win_all\Scripts\Activate.ps1
+python coderai
+```
+
 That's it. Open `http://127.0.0.1:8776/admin` and log in with `admin` / `admin`.
 
 ---
@@ -152,8 +168,67 @@ cd coderai
 ./build.sh vulkan   # AMD/Intel only
 ```
 
+Platform-specific alternatives:
+
+```bash
+./osxbuild.sh all   # macOS, prefers Metal-backed builds when available
+```
+
+```powershell
+.\build.ps1 -Backend all   # Windows, prefers CUDA-backed builds when available
+```
+
+Packaging options:
+
+```bash
+./build.sh all --package
+./osxbuild.sh all --package
+```
+
+```powershell
+.\build.ps1 -Backend all -Package
+```
+
+`--package` installs PyInstaller into the build virtual environment and produces a self-contained distributable from the venv that was just created or updated.
+
+Packaging outputs:
+- Linux: `dist-package/coderai`
+- macOS: `dist-package/coderai` and `dist-package/CoderAI.app`
+- Windows: `dist-package/coderai.exe`
+
+Packaging notes:
+- macOS does have an equivalent to a standalone packaged app: a `.app` bundle. `osxbuild.sh --package` now builds both a single CLI binary and a macOS app bundle.
+- These packages bundle the Python interpreter and Python modules from the venv, but they do not eliminate the need for compatible external GPU/runtime drivers on the target machine.
+- CUDA builds on Linux and Windows still require matching NVIDIA driver/runtime support on the destination system.
+- Metal builds on macOS still require a compatible macOS system with Metal support.
+
 The build script creates a virtual environment, installs dependencies, and builds GPU-accelerated backends including `stable-diffusion-cpp-python` with CUDA+Vulkan support.
 
+Platform backend notes:
+- Linux: CUDA for NVIDIA, Vulkan for AMD/Intel/NVIDIA, OpenCL fallback where supported.
+- macOS: Metal is the correct GPU acceleration path instead of CUDA. `osxbuild.sh` uses PyTorch MPS plus `GGML_METAL` / `SD_METAL` builds where available.
+- Windows: CUDA remains the primary NVIDIA acceleration path. `build.ps1` focuses on CUDA or CPU installs.
+- There is no general-purpose CUDA workflow for current macOS systems; Apple GPU acceleration uses Metal.
+
+### Platform Support Matrix
+
+| Capability | Linux | macOS | Windows |
+|---|---|---|---|
+| Core server / admin UI | Yes | Yes | Yes |
+| Default path handling | Yes | Yes | Yes |
+| PyTorch GPU acceleration | CUDA | Metal (MPS) | CUDA |
+| `llama-cpp-python` GPU path | CUDA / Vulkan | Metal | CUDA |
+| `stable-diffusion-cpp-python` GPU path | CUDA / Vulkan / OpenCL | Metal | CUDA |
+| `whisper.cpp` accelerated path | Vulkan / CPU fallback | Metal / CPU fallback | CPU fallback |
+| InsightFace / ONNX runtime | `onnxruntime-gpu` | `onnxruntime-silicon` or CPU | `onnxruntime-gpu` |
+| Build script included in repo | `build.sh` | `osxbuild.sh` | `build.ps1` |
+
+Notes:
+- "Yes" means CoderAI has an intended path for that platform, not that every optional dependency is guaranteed to install on every machine.
+- macOS GPU acceleration is Metal-based; there is no standard modern CUDA path for macOS.
+- Windows currently uses CUDA as the main NVIDIA acceleration path; Vulkan/OpenCL build flows are not the primary Windows setup in this repository.
+- Some optional audio and media packages may still vary by Python version, hardware, and upstream wheel availability.
+
 ### Manual Installation
 
 ```bash
@@ -247,6 +322,45 @@ Config files live in `~/.coderai/` (or `--config` path):
 └── secret_key       # Session signing key (auto-generated)
 ```
 
+### AISBF Broker Client
+
+CoderAI includes an AISBF broker websocket client that can register this instance
+with a broker and receive brokered requests.
+
+You can configure it either by editing `config.json` directly or from the admin
+Settings page under `AISBF Broker`.
+
+Example broker configuration:
+
+```json
+{
+  "broker": {
+    "enabled": true,
+    "base_url": "https://broker.example.com",
+    "scope": "user",
+    "username": "alice",
+    "provider_id": "coderai-local",
+    "client_id": "workstation-01",
+    "registration_token": "your-registration-token",
+    "advertised_endpoint": "http://127.0.0.1:8776",
+    "transport": "websocket",
+    "heartbeat_interval_seconds": 30,
+    "connect_timeout_seconds": 10,
+    "request_timeout_seconds": 30,
+    "reconnect_initial_delay_seconds": 1,
+    "reconnect_max_delay_seconds": 60
+  }
+}
+```
+
+Broker notes:
+- `base_url` accepts `http`, `https`, `ws`, or `wss`; the websocket route is derived automatically.
+- `scope: "user"` requires a non-global `username`.
+- `scope: "global"` requires `username: "global"`.
+- When `enabled` is `true`, `provider_id`, `client_id`, and `registration_token` are required.
+- `advertised_endpoint` is optional and is sent to the broker as the externally reachable endpoint for this instance.
+- Restart CoderAI after changing broker settings so the background broker service reconnects with the new configuration.
+
 ### config.json
 
 ```json

build.ps1

+param(
+    [string]$Backend = "all",
+    [string]$Venv = "",
+    [switch]$Package
+)
+
+$ErrorActionPreference = "Stop"
+$Backend = $Backend.ToLowerInvariant()
+$validBackends = @("cuda", "cpu", "all")
+if ($validBackends -notcontains $Backend) {
+    throw "Invalid backend '$Backend'. Use one of: cuda, cpu, all"
+}
+
+function Write-Info($msg) { Write-Host $msg -ForegroundColor Cyan }
+function Write-Warn($msg) { Write-Host $msg -ForegroundColor Yellow }
+function Write-Ok($msg) { Write-Host $msg -ForegroundColor Green }
+
+Write-Info "========================================"
+Write-Info "  CoderAI Windows Build Script"
+Write-Info "  Backend: $Backend"
+Write-Info "========================================"
+
+$python = if (Get-Command py -ErrorAction SilentlyContinue) { "py" } else { "python" }
+$pythonArgs = if ($python -eq "py") { @("-3") } else { @() }
+
+if ([string]::IsNullOrWhiteSpace($Venv)) {
+    switch ($Backend) {
+        "cuda" { $Venv = "venv_win_cuda" }
+        "cpu" { $Venv = "venv_win_cpu" }
+        default { $Venv = "venv_win_all" }
+    }
+}
+
+if (-not (Test-Path $Venv)) {
+    Write-Info "Creating virtual environment: $Venv"
+    & $python @pythonArgs -m venv $Venv
+} else {
+    Write-Warn "Using existing virtual environment: $Venv"
+}
+
+$venvPython = Join-Path $Venv "Scripts\python.exe"
+$venvActivate = Join-Path $Venv "Scripts\Activate.ps1"
+
+& $venvPython -m pip install --upgrade pip setuptools wheel
+& $venvPython -m pip install -r requirements.txt
+
+function Install-CommonMLStack {
+    & $venvPython -m pip install "imageio[ffmpeg]" scipy soundfile sentence-transformers openai-whisper argostranslate edge-tts kokoro-tts timm
+    & $venvPython -m pip install realesrgan basicsr
+    & $venvPython -m pip install demucs deepfilternet rnnoise voicefixer
+    & $venvPython -m pip install insightface onnxruntime-gpu
+    & $venvPython -m pip install f5-tts seed-vc
+    try { & $venvPython -m pip install audiocraft } catch { Write-Warn "audiocraft install failed; continuing" }
+}
+
+function Install-CudaStack {
+    Write-Info "Installing PyTorch with CUDA support..."
+    & $venvPython -m pip install torch torchvision torchaudio
+
+    Write-Info "Installing NVIDIA-specific requirements..."
+    try { & $venvPython -m pip install -r requirements-nvidia.txt } catch { Write-Warn "Some NVIDIA-specific packages failed; continuing" }
+
+    Write-Info "Installing llama-cpp-python with CUDA support..."
+    $env:CMAKE_ARGS = "-DGGML_CUDA=ON"
+    try { & $venvPython -m pip install --upgrade llama-cpp-python --no-cache-dir } finally { Remove-Item Env:CMAKE_ARGS -ErrorAction SilentlyContinue }
+
+    Write-Info "Installing stable-diffusion-cpp-python with CUDA support..."
+    $env:CMAKE_ARGS = "-DSD_CUDA=ON -DSD_WEBM=OFF"
+    try { & $venvPython -m pip install stable-diffusion-cpp-python --no-cache-dir } catch { Write-Warn "stable-diffusion-cpp-python CUDA build failed; continuing" } finally { Remove-Item Env:CMAKE_ARGS -ErrorAction SilentlyContinue }
+}
+
+function Install-CpuStack {
+    Write-Info "Installing CPU-oriented runtime..."
+    & $venvPython -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
+    & $venvPython -m pip install --upgrade llama-cpp-python
+    try { & $venvPython -m pip install stable-diffusion-cpp-python } catch { Write-Warn "stable-diffusion-cpp-python CPU install failed; continuing" }
+    try { & $venvPython -m pip install onnxruntime } catch { Write-Warn "onnxruntime CPU install failed; continuing" }
+}
+
+function Invoke-PackageBuild {
+    Write-Info "Packaging CoderAI with PyInstaller..."
+    & $venvPython -m pip install pyinstaller
+    New-Item -ItemType Directory -Force -Path "dist-package" | Out-Null
+    & $venvPython -m PyInstaller --clean --noconfirm --onefile --name coderai `
+        --collect-all codai `
+        --collect-all fastapi `
+        --collect-all uvicorn `
+        --collect-all pydantic `
+        --collect-all transformers `
+        --collect-all diffusers `
+        --collect-all sentence_transformers `
+        --collect-all whispercpp `
+        --collect-all insightface `
+        --collect-all onnxruntime `
+        --collect-all PIL `
+        coderai
+    Copy-Item "dist\coderai.exe" "dist-package\coderai.exe" -Force
+    Write-Ok "Packaged executable: dist-package\coderai.exe"
+    Write-Warn "Target systems still need compatible GPU/runtime drivers such as CUDA when GPU backends are used."
+}
+
+switch ($Backend) {
+    "cuda" {
+        Install-CudaStack
+        Install-CommonMLStack
+    }
+    "cpu" {
+        Install-CpuStack
+        Install-CommonMLStack
+    }
+    default {
+        Install-CudaStack
+        Install-CommonMLStack
+    }
+}
+
+Set-Content -Path ".backend" -Value $Backend
+if ($Package) {
+    Invoke-PackageBuild
+}
+Write-Ok "Build completed successfully!"
+Write-Host ""
+Write-Host "To activate the environment in the future, run:"
+Write-Host "  $venvActivate"
+Write-Host ""
+Write-Host "Recommended runtime notes:"
+Write-Host "  - Windows keeps CUDA as the primary NVIDIA acceleration path"
+Write-Host "  - Metal is macOS-only and not used on Windows"

build.sh

@@ -16,7 +16,7 @@
 # along with this program. If not, see <https://www.gnu.org/licenses/>.
 
 # Build script for CoderAI - Supports NVIDIA (CUDA), Vulkan, OpenCL, and CPU backends
-# Usage: ./build.sh [nvidia|vulkan|vulkan-nvidia|cuda|opencl|all] [--flash] [--venv <venv>]
+# Usage: ./build.sh [nvidia|vulkan|vulkan-nvidia|cuda|opencl|all] [--flash] [--venv <venv>] [--package]
 # Default: all (installs all backends)
 # --flash: Enable and install Flash Attention 2 (for NVIDIA GPUs)
 # --venv <venv>: Specify custom virtual environment name
@@ -34,6 +34,7 @@ NC='\033[0m' # No Color
 BACKEND="${1:-all}"
 FLASH=false
 CUSTOM_VENV=""
+PACKAGE=false
 
 # Parse arguments
 i=1
@@ -46,6 +47,9 @@ for arg in "$@"; do
             i=$((i + 1))
             eval "CUSTOM_VENV=\${$i}"
             ;;
+        --package)
+            PACKAGE=true
+            ;;
     esac
     i=$((i + 1))
 done
@@ -719,10 +723,36 @@ elif [ "$BACKEND" = "all" ]; then
     echo ""
 fi
 
+package_app() {
+    echo -e "${YELLOW}Packaging CoderAI with PyInstaller...${NC}"
+    pip install pyinstaller
+    mkdir -p dist-package
+    pyinstaller --clean --noconfirm --onefile --name coderai \
+        --collect-all codai \
+        --collect-all fastapi \
+        --collect-all uvicorn \
+        --collect-all pydantic \
+        --collect-all transformers \
+        --collect-all diffusers \
+        --collect-all sentence_transformers \
+        --collect-all whispercpp \
+        --collect-all insightface \
+        --collect-all onnxruntime \
+        --collect-all PIL \
+        coderai
+    cp dist/coderai dist-package/coderai
+    echo -e "${GREEN}✓ Packaged executable: dist-package/coderai${NC}"
+    echo -e "${YELLOW}Note: The target machine must still provide compatible system GPU/runtime libraries.${NC}"
+}
+
 # Create .backend file to track which backend was used
 echo "$BACKEND" > .backend
 
+if [ "$PACKAGE" = true ]; then
+    package_app
+fi
+
 echo -e "${GREEN}Build completed successfully!${NC}"
 echo ""
 echo "To activate the environment in the future, run:"
-echo "  source $VENV_DIR/bin/activate"
+echo "  source $VENV_DIR/bin/activate"
\ No newline at end of file

codai/admin/auth.py

@@ -54,7 +54,10 @@ def get_or_create_secret(config_dir: Path) -> bytes:
         secret = secrets.token_bytes(32)
         with open(secret_path, 'wb') as f:
             f.write(secret)
-        secret_path.chmod(0o600)
+        try:
+            secret_path.chmod(0o600)
+        except OSError:
+            pass
         return secret
 
 

codai/admin/routes.py

@@ -25,6 +25,7 @@ from fastapi.responses import HTMLResponse, RedirectResponse, JSONResponse, Stre
 from fastapi.templating import Jinja2Templates
 
 from codai.admin.auth import SessionManager
+from codai.platform_paths import default_whisper_server_path
 import queue as _q
 import threading as _t
 import uuid as _uuid
@@ -90,7 +91,7 @@ def _next_whisper_server_model_id(audio_models) -> str:
 
 
 def _default_whisper_server_path() -> str:
-    return shutil.which("whisper-server") or "/usr/local/bin/whisper-server"
+    return shutil.which("whisper-server") or default_whisper_server_path()
 
 
 def get_current_user(request: Request) -> Optional[str]:
@@ -239,7 +240,11 @@ async def admin_dashboard(request: Request, username: str = Depends(require_auth
 
 @router.get("/admin/models", response_class=HTMLResponse)
 async def models_page(request: Request, username: str = Depends(require_admin)):
-    return _tmpl(request, "models.html", {"username": username, "is_admin": True})
+    return _tmpl(request, "models.html", {
+        "username": username,
+        "is_admin": True,
+        "default_whisper_server_path": _default_whisper_server_path(),
+    })
 
 
 @router.get("/admin/tokens", response_class=HTMLResponse)
@@ -1434,12 +1439,14 @@ async def api_model_configure(request: Request, username: str = Depends(require_
         gpu_device = int(data.get("gpu_device", 0))
         if gpu_device < 0:
             raise HTTPException(status_code=400, detail="gpu_device must be >= 0")
-        # Remove existing entry with same id (update semantics)
         audio_list = config_manager.models_data.get("audio_models", [])
-        config_manager.models_data["audio_models"] = [
-            m for m in audio_list
-            if not (isinstance(m, dict) and m.get("id") == model_id)
-        ]
+        if any(
+            isinstance(m, dict)
+            and m.get("backend") == "whisper-server"
+            and m.get("id") == model_id
+            for m in audio_list
+        ):
+            raise HTTPException(status_code=409, detail=f"whisper-server model '{model_id}' already exists")
         alias = (data.get("alias") or "").strip() or None
         entry = {
             "id": model_id,
@@ -1617,6 +1624,22 @@ async def api_get_settings(username: str = Depends(require_admin)):
             "directory": c.archive.directory,
             "retention": c.archive.retention,
         },
+        "broker": {
+            "enabled": c.broker.enabled,
+            "base_url": c.broker.base_url,
+            "scope": c.broker.scope,
+            "username": c.broker.username,
+            "provider_id": c.broker.provider_id,
+            "client_id": c.broker.client_id,
+            "registration_token": c.broker.registration_token,
+            "advertised_endpoint": c.broker.advertised_endpoint,
+            "transport": c.broker.transport,
+            "heartbeat_interval_seconds": c.broker.heartbeat_interval_seconds,
+            "connect_timeout_seconds": c.broker.connect_timeout_seconds,
+            "request_timeout_seconds": c.broker.request_timeout_seconds,
+            "reconnect_initial_delay_seconds": c.broker.reconnect_initial_delay_seconds,
+            "reconnect_max_delay_seconds": c.broker.reconnect_max_delay_seconds,
+        },
         "system_prompt": c.system_prompt,
         "tools_closer_prompt": c.tools_closer_prompt,
         "grammar_guided": c.grammar_guided,
@@ -1707,6 +1730,28 @@ async def api_save_settings(request: Request, username: str = Depends(require_ad
         resolved = raw_dir if raw_dir and _os.path.isabs(raw_dir) else _os.path.join(cfg_dir, raw_dir or "archive")
         archive_manager.configure(c.archive.enabled, resolved, c.archive.retention)
 
+    if "broker" in data:
+        bro = data["broker"]
+        c.broker.enabled = bool(bro.get("enabled", c.broker.enabled))
+        c.broker.base_url = (bro.get("base_url") or "").strip()
+        c.broker.scope = (bro.get("scope") or c.broker.scope or "user").strip()
+        c.broker.username = (bro.get("username") or "").strip()
+        c.broker.provider_id = (bro.get("provider_id") or "").strip()
+        c.broker.client_id = (bro.get("client_id") or "").strip()
+        c.broker.registration_token = (bro.get("registration_token") or "").strip()
+        c.broker.advertised_endpoint = (bro.get("advertised_endpoint") or "").strip()
+        c.broker.transport = (bro.get("transport") or c.broker.transport or "websocket").strip()
+        c.broker.heartbeat_interval_seconds = max(1, int(bro.get("heartbeat_interval_seconds", c.broker.heartbeat_interval_seconds)))
+        c.broker.connect_timeout_seconds = max(1, int(bro.get("connect_timeout_seconds", c.broker.connect_timeout_seconds)))
+        c.broker.request_timeout_seconds = max(1, int(bro.get("request_timeout_seconds", c.broker.request_timeout_seconds)))
+        c.broker.reconnect_initial_delay_seconds = max(1, int(bro.get("reconnect_initial_delay_seconds", c.broker.reconnect_initial_delay_seconds)))
+        c.broker.reconnect_max_delay_seconds = max(
+            c.broker.reconnect_initial_delay_seconds,
+            int(bro.get("reconnect_max_delay_seconds", c.broker.reconnect_max_delay_seconds)),
+        )
+        from codai.broker.config import build_broker_runtime_config
+        request.app.state.broker_runtime = build_broker_runtime_config(c.broker)
+
     config_manager.save_config()
     return {"success": True}
 

codai/admin/templates/models.html

@@ -26,6 +26,9 @@
 {% endblock %}
 
 {% block content %}
+<script>
+window.__DEFAULT_WHISPER_SERVER_PATH__ = {{ default_whisper_server_path|tojson }};
+</script>
 <div class="page-header">
   <div>
     <h1>Models</h1>
@@ -1253,7 +1256,7 @@ function nextWhisperServerModelId(){
 }
 
 function defaultWhisperServerPath(){
-  return '/usr/local/bin/whisper-server';
+  return window.__DEFAULT_WHISPER_SERVER_PATH__ || '';
 }
 
 function resetWhisperServerBuilderDefaults(){

codai/admin/templates/settings.html

@@ -101,6 +101,81 @@
     <span class="form-hint">Archived entries older than this are automatically deleted. Takes effect immediately on save.</span>
   </div>
 </div>
+
+<div class="card mb-0" style="margin-top:1rem">
+  <div class="card-title">AISBF Broker</div>
+  <div class="form-row">
+    <label style="display:flex;align-items:center;gap:.5rem;cursor:pointer">
+      <input type="checkbox" id="s-broker-enabled" onchange="toggleBrokerFields()">
+      <span style="font-size:13px;font-weight:500">Enable broker client</span>
+    </label>
+    <span class="form-hint">Registers this CoderAI instance with an AISBF broker so it can receive brokered requests.</span>
+  </div>
+  <div id="broker-fields" style="display:none">
+    <div class="form-row">
+      <label class="form-label">Broker base URL</label>
+      <input type="text" id="s-broker-base-url" class="form-input" placeholder="https://broker.example.com">
+      <span class="form-hint">Supports `http`, `https`, `ws`, or `wss`. The websocket endpoint is derived automatically.</span>
+    </div>
+    <div style="display:grid;grid-template-columns:180px 1fr;gap:1rem;align-items:start">
+      <div class="form-row" style="margin:0">
+        <label class="form-label">Scope</label>
+        <select id="s-broker-scope" class="form-input">
+          <option value="user">user</option>
+          <option value="global">global</option>
+        </select>
+      </div>
+      <div class="form-row" style="margin:0">
+        <label class="form-label">Username</label>
+        <input type="text" id="s-broker-username" class="form-input" placeholder="alice or global">
+        <span class="form-hint">Use `global` when scope is `global`; otherwise provide the AISBF username.</span>
+      </div>
+    </div>
+    <div style="display:grid;grid-template-columns:1fr 1fr;gap:1rem;align-items:start">
+      <div class="form-row" style="margin:0">
+        <label class="form-label">Provider ID</label>
+        <input type="text" id="s-broker-provider-id" class="form-input" placeholder="coderai-local">
+      </div>
+      <div class="form-row" style="margin:0">
+        <label class="form-label">Client ID</label>
+        <input type="text" id="s-broker-client-id" class="form-input" placeholder="workstation-01">
+      </div>
+    </div>
+    <div class="form-row">
+      <label class="form-label">Registration token</label>
+      <input type="password" id="s-broker-registration-token" class="form-input" placeholder="Broker registration token">
+    </div>
+    <div class="form-row">
+      <label class="form-label">Advertised endpoint</label>
+      <input type="text" id="s-broker-advertised-endpoint" class="form-input" placeholder="http://127.0.0.1:8776">
+      <span class="form-hint">Optional external URL advertised to the broker for this instance.</span>
+    </div>
+    <div style="display:grid;grid-template-columns:repeat(3, minmax(0, 1fr));gap:1rem;align-items:start">
+      <div class="form-row" style="margin:0">
+        <label class="form-label">Heartbeat seconds</label>
+        <input type="number" id="s-broker-heartbeat" class="form-input" min="1" placeholder="30">
+      </div>
+      <div class="form-row" style="margin:0">
+        <label class="form-label">Connect timeout</label>
+        <input type="number" id="s-broker-connect-timeout" class="form-input" min="1" placeholder="10">
+      </div>
+      <div class="form-row" style="margin:0">
+        <label class="form-label">Request timeout</label>
+        <input type="number" id="s-broker-request-timeout" class="form-input" min="1" placeholder="30">
+      </div>
+    </div>
+    <div style="display:grid;grid-template-columns:repeat(2, minmax(0, 1fr));gap:1rem;align-items:start">
+      <div class="form-row" style="margin:0">
+        <label class="form-label">Reconnect initial delay</label>
+        <input type="number" id="s-broker-reconnect-initial" class="form-input" min="1" placeholder="1">
+      </div>
+      <div class="form-row" style="margin:0">
+        <label class="form-label">Reconnect max delay</label>
+        <input type="number" id="s-broker-reconnect-max" class="form-input" min="1" placeholder="60">
+      </div>
+    </div>
+  </div>
+</div>
 {% endblock %}
 
 {% block scripts %}
@@ -110,6 +185,11 @@ function toggleHttps(){
     document.getElementById('s-https').checked ? 'block' : 'none';
 }
 
+function toggleBrokerFields(){
+  document.getElementById('broker-fields').style.display =
+    document.getElementById('s-broker-enabled').checked ? 'block' : 'none';
+}
+
 function showAlert(type, msg){
   const el = document.getElementById('settings-alert');
   el.className = 'alert alert-' + (type === 'error' ? 'error' : 'info');
@@ -143,6 +223,21 @@ async function loadSettings(){
       if (hint && as.default_directory) hint.textContent = `(default: ${as.default_directory})`;
       document.getElementById('s-arc-dir').placeholder = as.default_directory || '(default)';
     } catch(_){}
+    const broker = d.broker || {};
+    document.getElementById('s-broker-enabled').checked = !!broker.enabled;
+    document.getElementById('s-broker-base-url').value = broker.base_url ?? '';
+    document.getElementById('s-broker-scope').value = broker.scope ?? 'user';
+    document.getElementById('s-broker-username').value = broker.username ?? '';
+    document.getElementById('s-broker-provider-id').value = broker.provider_id ?? '';
+    document.getElementById('s-broker-client-id').value = broker.client_id ?? '';
+    document.getElementById('s-broker-registration-token').value = broker.registration_token ?? '';
+    document.getElementById('s-broker-advertised-endpoint').value = broker.advertised_endpoint ?? '';
+    document.getElementById('s-broker-heartbeat').value = broker.heartbeat_interval_seconds ?? 30;
+    document.getElementById('s-broker-connect-timeout').value = broker.connect_timeout_seconds ?? 10;
+    document.getElementById('s-broker-request-timeout').value = broker.request_timeout_seconds ?? 30;
+    document.getElementById('s-broker-reconnect-initial').value = broker.reconnect_initial_delay_seconds ?? 1;
+    document.getElementById('s-broker-reconnect-max').value = broker.reconnect_max_delay_seconds ?? 60;
+    toggleBrokerFields();
   }catch(e){ showAlert('error','Failed to load settings: '+e.message); }
 }
 
@@ -169,6 +264,22 @@ async function saveSettings(){
       directory: document.getElementById('s-arc-dir').value.trim(),
       retention: document.getElementById('s-arc-retention').value,
     },
+    broker:{
+      enabled: document.getElementById('s-broker-enabled').checked,
+      base_url: document.getElementById('s-broker-base-url').value.trim(),
+      scope: document.getElementById('s-broker-scope').value,
+      username: document.getElementById('s-broker-username').value.trim(),
+      provider_id: document.getElementById('s-broker-provider-id').value.trim(),
+      client_id: document.getElementById('s-broker-client-id').value.trim(),
+      registration_token: document.getElementById('s-broker-registration-token').value.trim(),
+      advertised_endpoint: document.getElementById('s-broker-advertised-endpoint').value.trim(),
+      heartbeat_interval_seconds: parseInt(document.getElementById('s-broker-heartbeat').value) || 30,
+      connect_timeout_seconds: parseInt(document.getElementById('s-broker-connect-timeout').value) || 10,
+      request_timeout_seconds: parseInt(document.getElementById('s-broker-request-timeout').value) || 30,
+      reconnect_initial_delay_seconds: parseInt(document.getElementById('s-broker-reconnect-initial').value) || 1,
+      reconnect_max_delay_seconds: parseInt(document.getElementById('s-broker-reconnect-max').value) || 60,
+      transport: 'websocket',
+    },
   };
   try{
     const r = await fetch(ROOT_PATH + '/admin/api/settings',{

codai/api/characters.py

@@ -40,6 +40,8 @@ from typing import List, Optional
 from fastapi import APIRouter, HTTPException, Request
 from pydantic import BaseModel, ConfigDict
 
+from codai.platform_paths import default_characters_dir, legacy_style_config_dir
+
 router = APIRouter()
 
 _CHARS_DIR: Optional[str] = None
@@ -47,8 +49,8 @@ _CHARS_DIR: Optional[str] = None
 
 def set_global_args(args):
     global _CHARS_DIR
-    base = getattr(args, 'file_path', None) or os.path.expanduser('~/.coderai')
-    root = base if os.path.isdir(base) else (os.path.dirname(base) if base else os.path.expanduser('~/.coderai'))
+    base = getattr(args, 'file_path', None) or str(legacy_style_config_dir())
+    root = base if os.path.isdir(base) else (os.path.dirname(base) if base else str(legacy_style_config_dir()))
     _CHARS_DIR = os.path.join(root, 'characters')
     os.makedirs(_CHARS_DIR, exist_ok=True)
 
@@ -60,7 +62,7 @@ def set_global_file_path(path: str):
 def _chars_dir() -> str:
     if _CHARS_DIR:
         return _CHARS_DIR
-    d = os.path.expanduser('~/.coderai/characters')
+    d = str(default_characters_dir())
     os.makedirs(d, exist_ok=True)
     return d
 

codai/api/environments.py

@@ -42,6 +42,8 @@ from typing import List, Optional
 from fastapi import APIRouter, HTTPException, Request
 from pydantic import BaseModel, ConfigDict
 
+from codai.platform_paths import default_environments_dir, legacy_style_config_dir
+
 router = APIRouter()
 
 _ENVS_DIR: Optional[str] = None
@@ -49,8 +51,8 @@ _ENVS_DIR: Optional[str] = None
 
 def set_global_args(args):
     global _ENVS_DIR
-    base = getattr(args, 'file_path', None) or os.path.expanduser('~/.coderai')
-    root = base if os.path.isdir(base) else (os.path.dirname(base) if base else os.path.expanduser('~/.coderai'))
+    base = getattr(args, 'file_path', None) or str(legacy_style_config_dir())
+    root = base if os.path.isdir(base) else (os.path.dirname(base) if base else str(legacy_style_config_dir()))
     _ENVS_DIR = os.path.join(root, 'environments')
     os.makedirs(_ENVS_DIR, exist_ok=True)
 
@@ -62,7 +64,7 @@ def set_global_file_path(path: str):
 def _envs_dir() -> str:
     if _ENVS_DIR:
         return _ENVS_DIR
-    d = os.path.expanduser('~/.coderai/environments')
+    d = str(default_environments_dir())
     os.makedirs(d, exist_ok=True)
     return d
 

codai/api/faceswap.py

@@ -20,13 +20,14 @@ from PIL import Image
 from pydantic import BaseModel, ConfigDict
 
 from codai.api.images import save_image_response
+from codai.platform_paths import default_insightface_model_path
 
 router = APIRouter()
 
 global_args = None
 global_file_path = None
 
-_INSWAPPER_MODEL_PATH = os.path.expanduser('~/.insightface/models/inswapper_128.onnx')
+_INSWAPPER_MODEL_PATH = str(default_insightface_model_path())
 _INSWAPPER_HF_REPO = 'deepinsight/inswapper'
 _INSWAPPER_HF_FILE = 'inswapper_128.onnx'
 

codai/api/images.py

@@ -1713,7 +1713,8 @@ def _run_unpixelate(image_bytes: bytes, scale: int, model_path: Optional[str]) -
         mp = model_path
     else:
         # Download RealESRGAN_x4plus on demand
-        mp = os.path.expanduser('~/.cache/realesrgan/RealESRGAN_x4plus.pth')
+        from codai.platform_paths import default_realesrgan_model_path
+        mp = str(default_realesrgan_model_path())
         if not os.path.exists(mp):
             os.makedirs(os.path.dirname(mp), exist_ok=True)
             import urllib.request

codai/api/transcriptions.py

@@ -136,9 +136,11 @@ async def create_transcription(
 
     # Check if the requested model maps to a configured whisper-server instance first.
     # Try alias round-robin resolution before direct ID lookup.
+    whisper_model_id = multi_model_manager.resolve_whisper_alias_model_id(model)
     whisper_server = (
-        multi_model_manager.resolve_whisper_alias(model)
-        or multi_model_manager.whisper_servers.get(model)
+        multi_model_manager.whisper_servers.get(whisper_model_id)
+        if whisper_model_id is not None
+        else multi_model_manager.whisper_servers.get(model)
     )
     if whisper_server is not None:
         multi_model_manager.request_model(requested_model=model, model_type="audio")
@@ -148,7 +150,7 @@ async def create_transcription(
                 gpu_device=getattr(whisper_server, "_gpu_device", 0),
             )
             if whisper_server.is_running():
-                ws_key = f"audio:{model}"
+                ws_key = f"audio:{whisper_model_id or model}"
                 multi_model_manager.models[ws_key] = whisper_server
                 multi_model_manager.active_in_vram = ws_key
                 multi_model_manager.models_in_vram.add(ws_key)

codai/api/voice_clone.py

@@ -22,6 +22,8 @@ from typing import Optional
 from fastapi import APIRouter, HTTPException, Request, UploadFile, File, Form
 from pydantic import BaseModel, ConfigDict
 
+from codai.platform_paths import default_voices_dir
+
 router = APIRouter()
 
 global_args = None
@@ -35,8 +37,8 @@ def set_global_args(args):
     global global_args, _VOICES_DIR
     global_args = args
     # Store voice profiles alongside output files, or in a default location
-    base = getattr(args, 'file_path', None) or os.path.expanduser('~/.coderai/voices')
-    _VOICES_DIR = os.path.join(base if os.path.isdir(base) else os.path.dirname(base) if base else os.path.expanduser('~/.coderai'), 'voices')
+    base = getattr(args, 'file_path', None) or str(default_voices_dir())
+    _VOICES_DIR = os.path.join(base if os.path.isdir(base) else os.path.dirname(base) if base else str(default_voices_dir().parent), 'voices')
     os.makedirs(_VOICES_DIR, exist_ok=True)
 
 
@@ -48,7 +50,7 @@ def set_global_file_path(path):
 def _voices_dir() -> str:
     if _VOICES_DIR:
         return _VOICES_DIR
-    d = os.path.expanduser('~/.coderai/voices')
+    d = str(default_voices_dir())
     os.makedirs(d, exist_ok=True)
     return d
 

codai/cli.py

@@ -20,6 +20,8 @@ import json
 import os
 from pathlib import Path
 
+from codai.platform_paths import legacy_style_config_dir
+
 
 def load_config_file(config_dir: Path) -> dict:
     """Load the main config.json file."""
@@ -166,11 +168,12 @@ def setup_default_config(config_dir: Path):
 
 def parse_args():
     """Parse command line arguments."""
+    default_config = str(legacy_style_config_dir())
     parser = argparse.ArgumentParser(
         description="OpenAI-compatible API server supporting NVIDIA (CUDA) and Vulkan backends",
         formatter_class=argparse.RawDescriptionHelpFormatter,
         epilog="""Configuration: All settings are loaded from JSON config files in the
-configuration directory (--config DIR, default: ~/.coderai/). Key files:
+configuration directory (--config DIR, default: OS-specific CoderAI directory). Key files:
   config.json  - Server and backend settings
   models.json  - Model registry and configurations
   auth.json    - Users, tokens, and sessions"""
@@ -178,8 +181,8 @@ configuration directory (--config DIR, default: ~/.coderai/). Key files:
     parser.add_argument(
         "--config",
         type=str,
-        default=os.path.expanduser("~/.coderai/"),
-        help="Configuration directory (default: ~/.coderai/)",
+        default=default_config,
+        help=f"Configuration directory (default: {default_config})",
     )
     parser.add_argument(
         "--debug",
@@ -224,4 +227,4 @@ configuration directory (--config DIR, default: ~/.coderai/). Key files:
         action="store_true",
         help="List available Vulkan GPU devices and exit",
     )
-    return parser.parse_args()
+    return parser.parse_args()
\ No newline at end of file

codai/models/cache/__init__.py

@@ -40,6 +40,8 @@ import hashlib
 import pathlib
 from typing import Optional, Dict, List, Tuple
 
+from codai.platform_paths import default_model_cache_dir, legacy_style_cache_root
+
 # For type hints
 import time
 
@@ -49,8 +51,7 @@ def get_model_cache_dir() -> str:
     if os.environ.get('CODERAI_CACHE_DIR'):
         cache_dir = os.environ['CODERAI_CACHE_DIR']
     else:
-        cache_home = os.environ.get('XDG_CACHE_HOME', os.path.expanduser('~/.cache'))
-        cache_dir = os.path.join(cache_home, 'coderai', 'models')
+        cache_dir = str(default_model_cache_dir())
     pathlib.Path(cache_dir).mkdir(parents=True, exist_ok=True)
     return cache_dir
 
@@ -58,7 +59,7 @@ def get_model_cache_dir() -> str:
 def get_all_cache_dirs() -> dict:
     """Get all model cache directories."""
     caches = {}
-    cache_home = os.environ.get('XDG_CACHE_HOME', os.path.expanduser('~/.cache'))
+    cache_home = str(legacy_style_cache_root())
 
     # Coderai GGUF cache
     coderai_cache = get_model_cache_dir()
@@ -79,7 +80,7 @@ def get_all_cache_dirs() -> dict:
             caches['huggingface'] = hf_cache
 
     # Local diffusers cache
-    local_diffusers = os.path.expanduser('~/.cache/diffusers')
+    local_diffusers = os.path.join(cache_home, 'diffusers')
     if os.path.exists(local_diffusers):
         caches['diffusers'] = local_diffusers
 
@@ -549,4 +550,4 @@ __all__ = [
     'remove_cached_model',
     'list_cached_models_info',
     'remove_all_cached_models',
-]
+]
\ No newline at end of file

codai/models/manager.py

@@ -18,6 +18,7 @@
 
 from typing import Optional, Dict, Any, List
 import os
+import random
 import subprocess
 import signal
 import requests
@@ -256,6 +257,7 @@ class WhisperServerManager:
         self.current_model = None
         self.base_url = f"http://127.0.0.1:{port}"
         self.lock = threading.Lock()
+        self._active_requests = 0
         
         # Check if port is available
         if not self._is_port_available(port):
@@ -351,7 +353,8 @@ class WhisperServerManager:
         """Send transcription request to whisper-server."""
         if not self.is_running():
             return {"error": "whisper-server not running"}
-        
+        with self.lock:
+            self._active_requests += 1
         try:
             files = {"file": ("audio.wav", audio_data, "audio/wav")}
             data = {}
@@ -373,6 +376,16 @@ class WhisperServerManager:
                 return {"error": f"Server error: {response.status_code}", "detail": response.text}
         except Exception as e:
             return {"error": str(e)}
+        finally:
+            with self.lock:
+                self._active_requests = max(0, self._active_requests - 1)
+
+    def active_requests(self) -> int:
+        with self.lock:
+            return self._active_requests
+
+    def is_idle(self) -> bool:
+        return self.active_requests() == 0
     
     def _wait_for_server(self, timeout: int = 30) -> bool:
         """Wait for whisper-server to be ready."""
@@ -810,6 +823,17 @@ class MultiModelManager:
         idx = self._whisper_alias_counters.get(name, 0) % len(ids)
         self._whisper_alias_counters[name] = idx + 1
         return self.whisper_servers.get(ids[idx])
+
+    def resolve_whisper_alias_model_id(self, name: str) -> Optional[str]:
+        """Return an idle whisper-server model id for an alias, else a random one."""
+        ids = self.whisper_aliases.get(name)
+        if not ids:
+            return None
+        idle_ids = [mid for mid in ids if (self.whisper_servers.get(mid) and self.whisper_servers[mid].is_idle())]
+        choices = idle_ids or list(ids)
+        if len(choices) == 1:
+            return choices[0]
+        return random.choice(choices)
     
     def set_tts_model(self, model_name: str, config: Dict = None):
         """Set the text-to-speech model and download/cache it if needed."""
@@ -900,6 +924,9 @@ class MultiModelManager:
             for m in self.audio_models:
                 allowed.add(m)
                 allowed.add(f"audio:{m}")
+        for alias in self.whisper_aliases:
+            allowed.add(alias)
+            allowed.add(f"audio:{alias}")
 
         # TTS model
         if self.tts_model:
@@ -963,10 +990,18 @@ class MultiModelManager:
                             "video_models", "audio_gen_models", "embedding_models",
                             "spatial_models"):
                     for m in md.get(cat, []):
-                        mid = (m if isinstance(m, str) else
-                               m.get("alias") or m.get("path") or m.get("id") or "")
+                        if isinstance(m, str):
+                            mid = m
+                        elif m.get("backend") == "whisper-server":
+                            mid = m.get("id") or ""
+                        else:
+                            mid = m.get("alias") or m.get("path") or m.get("id") or ""
                         raw = (m if isinstance(m, str) else m.get("path") or m.get("id") or "")
-                        for val in (mid, raw):
+                        alias = "" if isinstance(m, str) else (m.get("alias") or "")
+                        vals = [mid, raw]
+                        if alias:
+                            vals.append(alias)
+                        for val in vals:
                             if val:
                                 allowed.add(val)
                                 short = val.split("/")[-1] if "/" in val else val

codai/platform_paths.py

+"""Platform-aware filesystem helpers for CoderAI."""
+
+from __future__ import annotations
+
+import os
+from pathlib import Path
+
+
+APP_NAME = "coderai"
+
+
+def _home_dir() -> Path:
+    return Path.home()
+
+
+def _windows_dir(env_var: str, fallback: Path) -> Path:
+    value = os.environ.get(env_var)
+    return Path(value).expanduser() if value else fallback
+
+
+def user_config_dir() -> Path:
+    if os.name == "nt":
+        base = _windows_dir("APPDATA", _home_dir() / "AppData" / "Roaming")
+        return base / "CoderAI"
+    if sys_platform() == "darwin":
+        return _home_dir() / "Library" / "Application Support" / "CoderAI"
+    xdg = os.environ.get("XDG_CONFIG_HOME")
+    base = Path(xdg).expanduser() if xdg else (_home_dir() / ".config")
+    return base / APP_NAME
+
+
+def user_data_dir() -> Path:
+    if os.name == "nt":
+        base = _windows_dir("LOCALAPPDATA", _home_dir() / "AppData" / "Local")
+        return base / "CoderAI"
+    if sys_platform() == "darwin":
+        return _home_dir() / "Library" / "Application Support" / "CoderAI"
+    xdg = os.environ.get("XDG_DATA_HOME")
+    base = Path(xdg).expanduser() if xdg else (_home_dir() / ".local" / "share")
+    return base / APP_NAME
+
+
+def user_cache_dir() -> Path:
+    if os.name == "nt":
+        base = _windows_dir("LOCALAPPDATA", _home_dir() / "AppData" / "Local")
+        return base / "CoderAI" / "Cache"
+    if sys_platform() == "darwin":
+        return _home_dir() / "Library" / "Caches" / "CoderAI"
+    xdg = os.environ.get("XDG_CACHE_HOME")
+    base = Path(xdg).expanduser() if xdg else (_home_dir() / ".cache")
+    return base / APP_NAME
+
+
+def ensure_dir(path: Path) -> Path:
+    path.mkdir(parents=True, exist_ok=True)
+    return path
+
+
+def default_config_dir() -> Path:
+    return ensure_dir(user_config_dir())
+
+
+def default_data_dir() -> Path:
+    return ensure_dir(user_data_dir())
+
+
+def default_cache_dir() -> Path:
+    return ensure_dir(user_cache_dir())
+
+
+def legacy_style_config_dir() -> Path:
+    if os.name == "nt":
+        return _home_dir() / ".coderai"
+    return _home_dir() / ".coderai"
+
+
+def legacy_style_cache_root() -> Path:
+    if os.name == "nt":
+        base = _windows_dir("LOCALAPPDATA", _home_dir() / "AppData" / "Local")
+        return base / ".cache"
+    if sys_platform() == "darwin":
+        return _home_dir() / "Library" / "Caches"
+    xdg = os.environ.get("XDG_CACHE_HOME")
+    return Path(xdg).expanduser() if xdg else (_home_dir() / ".cache")
+
+
+def default_model_cache_dir() -> Path:
+    return ensure_dir(legacy_style_cache_root() / APP_NAME / "models")
+
+
+def default_diffusers_cache_dir() -> Path:
+    return ensure_dir(legacy_style_cache_root() / "diffusers")
+
+
+def default_realesrgan_model_path() -> Path:
+    return default_diffusers_cache_dir().parent / "realesrgan" / "RealESRGAN_x4plus.pth"
+
+
+def default_insightface_model_path() -> Path:
+    return ensure_dir(legacy_style_cache_root() / "insightface" / "models") / "inswapper_128.onnx"
+
+
+def default_voices_dir() -> Path:
+    return ensure_dir(legacy_style_config_dir() / "voices")
+
+
+def default_characters_dir() -> Path:
+    return ensure_dir(legacy_style_config_dir() / "characters")
+
+
+def default_environments_dir() -> Path:
+    return ensure_dir(legacy_style_config_dir() / "environments")
+
+
+def default_whisper_server_path() -> str:
+    if os.name == "nt":
+        local = _windows_dir("LOCALAPPDATA", _home_dir() / "AppData" / "Local")
+        return str(local / "Programs" / "whisper-server" / "whisper-server.exe")
+    if sys_platform() == "darwin":
+        return "/usr/local/bin/whisper-server"
+    return "/usr/local/bin/whisper-server"
+
+
+def sys_platform() -> str:
+    return os.sys.platform

osxbuild.sh

+#!/bin/bash
+# CoderAI macOS build script
+# Usage: ./osxbuild.sh [metal|cpu|all] [--venv <venv>] [--package]
+
+set -e
+
+RED='\033[0;31m'
+GREEN='\033[0;32m'
+YELLOW='\033[1;33m'
+BLUE='\033[0;34m'
+NC='\033[0m'
+
+BACKEND="${1:-all}"
+CUSTOM_VENV=""
+PACKAGE=false
+
+i=1
+for arg in "$@"; do
+    case $arg in
+        --venv)
+            i=$((i + 1))
+            eval "CUSTOM_VENV=\${$i}"
+            ;;
+        --package)
+            PACKAGE=true
+            ;;
+    esac
+    i=$((i + 1))
+done
+
+BACKEND=$(echo "$BACKEND" | tr '[:upper:]' '[:lower:]')
+if [[ "$BACKEND" != "metal" && "$BACKEND" != "cpu" && "$BACKEND" != "all" ]]; then
+    echo -e "${RED}Error: Invalid backend '$BACKEND'${NC}"
+    echo "Usage: ./osxbuild.sh [metal|cpu|all] [--venv <venv>]"
+    exit 1
+fi
+
+echo -e "${BLUE}========================================${NC}"
+echo -e "${BLUE}  CoderAI macOS Build Script${NC}"
+echo -e "${BLUE}  Backend: ${GREEN}$BACKEND${NC}"
+echo -e "${BLUE}========================================${NC}"
+echo ""
+
+PYTHON_BIN="${PYTHON_BIN:-python3}"
+PYTHON_VERSION=$($PYTHON_BIN --version 2>&1 | python3 -c 'import sys,re; print(re.search(r"(\d+\.\d+)", sys.stdin.read()).group(1))')
+echo -e "${GREEN}✓ Python version: $PYTHON_VERSION${NC}"
+
+if [ -n "$CUSTOM_VENV" ]; then
+    VENV_DIR="$CUSTOM_VENV"
+elif [ "$BACKEND" = "metal" ]; then
+    VENV_DIR="venv_osx_metal"
+elif [ "$BACKEND" = "cpu" ]; then
+    VENV_DIR="venv_osx_cpu"
+else
+    VENV_DIR="venv_osx_all"
+fi
+
+if [ ! -d "$VENV_DIR" ]; then
+    echo -e "${YELLOW}Creating virtual environment: $VENV_DIR${NC}"
+    "$PYTHON_BIN" -m venv "$VENV_DIR"
+else
+    echo -e "${YELLOW}Using existing virtual environment: $VENV_DIR${NC}"
+fi
+
+source "$VENV_DIR/bin/activate"
+export PIP_NO_INPUT=1
+export PIP_REQUIRE_VIRTUALENV=1
+
+python -m pip install --upgrade pip setuptools wheel
+python -m pip install -r requirements.txt
+
+install_common_ml_stack() {
+    python -m pip install "imageio[ffmpeg]" scipy soundfile sentence-transformers openai-whisper argostranslate edge-tts kokoro-tts timm || true
+    python -m pip install realesrgan basicsr || true
+    python -m pip install demucs deepfilternet rnnoise voicefixer || true
+    python -m pip install insightface onnxruntime-silicon || python -m pip install insightface onnxruntime || true
+    python -m pip install f5-tts seed-vc || true
+    python -m pip install audiocraft || true
+}
+
+install_metal_stack() {
+    echo -e "${YELLOW}Installing PyTorch with Apple Silicon / MPS support...${NC}"
+    python -m pip install torch torchvision torchaudio
+
+    echo -e "${YELLOW}Installing llama-cpp-python with Metal support...${NC}"
+    CMAKE_ARGS="-DGGML_METAL=ON" python -m pip install --upgrade llama-cpp-python --no-cache-dir || {
+        echo -e "${YELLOW}Warning: Metal build failed, installing CPU llama-cpp-python${NC}"
+        python -m pip install --upgrade llama-cpp-python
+    }
+
+    echo -e "${YELLOW}Installing stable-diffusion-cpp-python with Metal support...${NC}"
+    CMAKE_ARGS="-DSD_METAL=ON -DSD_WEBM=OFF" python -m pip install stable-diffusion-cpp-python --no-cache-dir || {
+        echo -e "${YELLOW}Warning: Metal stable-diffusion-cpp-python not available${NC}"
+    }
+
+    echo -e "${YELLOW}Installing whispercpp with Metal support when possible...${NC}"
+    python -m pip uninstall -y whispercpp >/dev/null 2>&1 || true
+    TMP_DIR="${TMPDIR:-/tmp}/coderai-whispercpp"
+    rm -rf "$TMP_DIR"
+    git clone --depth 1 https://github.com/ggerganov/whisper.cpp "$TMP_DIR" >/dev/null 2>&1 || true
+    if [ -d "$TMP_DIR/bindings/python" ]; then
+        (cd "$TMP_DIR/bindings/python" && CMAKE_ARGS="-DWHISPER_METAL=ON -DGGML_METAL=ON" python -m pip install . --no-cache-dir --force-reinstall) || \
+        (cd "$TMP_DIR/bindings/python" && python -m pip install . --no-cache-dir --force-reinstall) || true
+    fi
+}
+
+install_cpu_stack() {
+    echo -e "${YELLOW}Installing CPU-only runtime...${NC}"
+    python -m pip install torch torchvision torchaudio
+    python -m pip install --upgrade llama-cpp-python
+    python -m pip install stable-diffusion-cpp-python || true
+}
+
+package_app() {
+    echo -e "${YELLOW}Packaging CoderAI with PyInstaller...${NC}"
+    python -m pip install pyinstaller
+    mkdir -p dist-package
+    pyinstaller --clean --noconfirm --onefile --name coderai \
+        --collect-all codai \
+        --collect-all fastapi \
+        --collect-all uvicorn \
+        --collect-all pydantic \
+        --collect-all transformers \
+        --collect-all diffusers \
+        --collect-all sentence_transformers \
+        --collect-all whispercpp \
+        --collect-all insightface \
+        --collect-all onnxruntime \
+        --collect-all PIL \
+        coderai
+    pyinstaller --clean --noconfirm --windowed --name CoderAI \
+        --collect-all codai \
+        --collect-all fastapi \
+        --collect-all uvicorn \
+        --collect-all pydantic \
+        --collect-all transformers \
+        --collect-all diffusers \
+        --collect-all sentence_transformers \
+        --collect-all whispercpp \
+        --collect-all insightface \
+        --collect-all onnxruntime \
+        --collect-all PIL \
+        coderai
+    cp dist/coderai dist-package/coderai
+    if [ -d "dist/CoderAI.app" ]; then
+        rm -rf dist-package/CoderAI.app
+        cp -R dist/CoderAI.app dist-package/CoderAI.app
+    fi
+    echo -e "${GREEN}✓ Packaged CLI binary: dist-package/coderai${NC}"
+    echo -e "${GREEN}✓ Packaged macOS app bundle: dist-package/CoderAI.app${NC}"
+    echo -e "${YELLOW}Note: macOS equivalent packaging is a single CLI binary plus a .app bundle; target machines still need compatible GPU/runtime libraries.${NC}"
+}
+
+if [ "$BACKEND" = "metal" ]; then
+    install_metal_stack
+    install_common_ml_stack
+elif [ "$BACKEND" = "cpu" ]; then
+    install_cpu_stack
+    install_common_ml_stack
+else
+    install_metal_stack
+    install_common_ml_stack
+fi
+
+echo "$BACKEND" > .backend
+if [ "$PACKAGE" = true ]; then
+    package_app
+fi
+echo -e "${GREEN}Build completed successfully!${NC}"
+echo ""
+echo "To activate the environment in the future, run:"
+echo "  source $VENV_DIR/bin/activate"
+echo ""
+echo "Recommended runtime notes:"
+echo "  - macOS uses Metal (MPS / GGML_METAL / SD_METAL) instead of CUDA"
+echo "  - NVIDIA CUDA is not the standard macOS path"

tests/test_whisper_server_local_models.py

@@ -247,6 +247,7 @@ def test_model_configure_allows_whisper_server_id_coexistence_with_other_audio_b
 
 def test_model_configure_defaults_missing_whisper_server_path_to_usr_local_bin(monkeypatch, tmp_path):
     from codai.admin import routes
+    from codai.platform_paths import default_whisper_server_path
 
     cfg = _build_config(tmp_path)
     monkeypatch.setattr(routes, "config_manager", cfg, raising=False)
@@ -267,7 +268,7 @@ def test_model_configure_defaults_missing_whisper_server_path_to_usr_local_bin(m
     )
 
     assert response.status_code == 200
-    assert cfg.models_data["audio_models"][0]["server_path"] == "/usr/local/bin/whisper-server"
+    assert cfg.models_data["audio_models"][0]["server_path"] == default_whisper_server_path()
 
     app.dependency_overrides.clear()
 
@@ -546,6 +547,67 @@ def test_transcription_requires_configured_whisper_server_model_id():
     assert "not configured" in str(exc.value.detail).lower() or "not available" in str(exc.value.detail).lower()
 
 
+def test_whisper_alias_prefers_idle_instance(monkeypatch):
+    from codai.models.manager import MultiModelManager
+
+    manager = MultiModelManager()
+    busy = SimpleNamespace(is_idle=lambda: False)
+    idle = SimpleNamespace(is_idle=lambda: True)
+    manager.whisper_servers = {"whisper0": busy, "whisper1": idle}
+    manager.whisper_aliases = {"shared-whisper": ["whisper0", "whisper1"]}
+
+    assert manager.resolve_whisper_alias_model_id("shared-whisper") == "whisper1"
+
+
+def test_whisper_alias_uses_random_choice_when_multiple_idle(monkeypatch):
+    from codai.models.manager import MultiModelManager
+
+    manager = MultiModelManager()
+    idle0 = SimpleNamespace(is_idle=lambda: True)
+    idle1 = SimpleNamespace(is_idle=lambda: True)
+    manager.whisper_servers = {"whisper0": idle0, "whisper1": idle1}
+    manager.whisper_aliases = {"shared-whisper": ["whisper0", "whisper1"]}
+    monkeypatch.setattr("codai.models.manager.random.choice", lambda items: items[-1])
+
+    assert manager.resolve_whisper_alias_model_id("shared-whisper") == "whisper1"
+
+
+def test_get_all_allowed_identifiers_includes_whisper_alias(monkeypatch):
+    from codai.admin import routes
+    from codai.models.manager import MultiModelManager
+
+    manager = MultiModelManager()
+    manager.audio_models[:] = ["whisper0", "whisper1"]
+    manager.whisper_aliases = {"shared-whisper": ["whisper0", "whisper1"]}
+    monkeypatch.setattr(
+        routes,
+        "config_manager",
+        SimpleNamespace(
+            models_data={
+                "text_models": [],
+                "image_models": [],
+                "audio_models": [
+                    {"id": "whisper0", "backend": "whisper-server", "alias": "shared-whisper"},
+                    {"id": "whisper1", "backend": "whisper-server", "alias": "shared-whisper"},
+                ],
+                "vision_models": [],
+                "tts_models": [],
+                "gguf_models": [],
+                "video_models": [],
+                "audio_gen_models": [],
+                "embedding_models": [],
+                "aliases": {},
+            }
+        ),
+        raising=False,
+    )
+
+    allowed = manager.get_all_allowed_identifiers()
+
+    assert "shared-whisper" in allowed
+    assert "audio:shared-whisper" in allowed
+
+
 def test_get_all_allowed_identifiers_includes_configured_whisper_server_id_without_legacy_alias(monkeypatch):
     from codai.admin import routes
     from codai.models.manager import MultiModelManager
@@ -673,7 +735,7 @@ def test_models_template_defines_whisper_server_builder_default_helpers():
     template = Path("codai/admin/templates/models.html").read_text()
 
     assert "function defaultWhisperServerPath()" in template
-    assert "return '/usr/local/bin/whisper-server';" in template
+    assert "return window.__DEFAULT_WHISPER_SERVER_PATH__ || '';" in template
     assert "function resetWhisperServerBuilderDefaults()" in template
     assert "const modelIdInput = document.getElementById('ws-model-id');" in template
     assert "const serverPathInput = document.getElementById('ws-server-path');" in template
@@ -779,3 +841,15 @@ def test_settings_template_keeps_queue_size_control():
 
     assert "Request queue max size" in template
     assert "s-queue-max" in template
+
+
+def test_settings_template_includes_broker_controls():
+    template = Path("codai/admin/templates/settings.html").read_text()
+
+    assert "AISBF Broker" in template
+    assert "s-broker-enabled" in template
+    assert "s-broker-base-url" in template
+    assert "s-broker-provider-id" in template
+    assert "s-broker-client-id" in template
+    assert "s-broker-registration-token" in template
+    assert "toggleBrokerFields()" in template