feat: manage whisper-server models from local models page

codai/admin/templates/models.html

@@ -94,6 +94,29 @@
     <div class="card-title">GGUF files <span id="gguf-file-badge" class="muted small"></span></div>
     <div id="gguf-models-list"><span class="muted small">Loading…</span></div>
   </div>
+
+  <div class="card mb-0" style="margin-top:1rem" id="ws-model-builder">
+    <div class="card-title">Whisper-server simulated models</div>
+    <p class="muted small" style="margin-top:0">Create local audio models backed by dedicated whisper-server subprocess configurations.</p>
+    <div style="display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:.75rem">
+      <input id="ws-model-id" class="form-input" placeholder="whisper-vulkan-base">
+      <input id="ws-server-path" class="form-input" placeholder="/usr/local/bin/whisper-server">
+      <input id="ws-model-path" class="form-input" placeholder="/models/ggml-base.bin">
+      <input id="ws-port" class="form-input" type="number" value="8744" min="1" max="65535">
+      <input id="ws-gpu-device" class="form-input" type="number" value="0" min="0">
+      <select id="ws-load-mode" class="form-input">
+        <option value="on-request">On request</option>
+        <option value="load">Load</option>
+      </select>
+    </div>
+    <div style="display:grid;grid-template-columns:repeat(2,minmax(0,1fr));gap:.75rem;margin-top:.75rem">
+      <input id="ws-used-vram" class="form-input" type="number" min="0" step="0.1" placeholder="Used VRAM (optional)">
+      <div></div>
+    </div>
+    <div class="form-actions" style="margin-top:.75rem">
+      <button class="btn btn-primary" onclick="addWhisperServerModel()">Add model</button>
+    </div>
+  </div>
 </div>
 
 <!-- SEARCH -->
@@ -939,6 +962,57 @@ async function loadCacheStats(){
 
 let _localModels = [];
 
+function _renderWhisperServerRows(models){
+  if(!models.length) return '';
+  const rows = models.map(m=>{
+    const idx = _localModels.length;
+    _localModels.push({
+      label:m.id,
+      path:m.id,
+      cacheType:'whisper-server',
+      size_gb:0,
+      defaultType:'audio_models',
+      settings:{
+        backend:m.backend || 'whisper-server',
+        load_mode:m.load_mode || 'on-request',
+        model_type:'audio_models',
+        model_path:m.model_path || '',
+        port:m.port,
+        gpu_device:m.gpu_device,
+      },
+      in_config:true,
+      capabilities:m.capabilities || ['speech_to_text']
+    });
+    const loaded = _loadedKeys.has(`audio:${m.id}`) || _loadedKeys.has(m.id);
+    return `<tr style="border-top:1px solid var(--border)">
+      <td style="padding:.4rem .25rem;font-family:monospace;font-size:12px">${esc(m.id)}</td>
+      <td style="padding:.4rem .25rem"><span class="badge badge-ok">${esc(m.backend || 'whisper-server')}</span></td>
+      <td style="padding:.4rem .25rem;font-size:11px;color:var(--text-2)">${esc(m.model_path || '—')}</td>
+      <td style="padding:.4rem .25rem;font-size:11px;color:var(--text-2)">${m.port ?? '—'} / GPU ${m.gpu_device ?? 0}</td>
+      <td style="padding:.4rem .25rem;font-size:11px;color:var(--text-2)">${esc(m.load_mode || 'on-request')}</td>
+      <td style="padding:.4rem .25rem;text-align:center">${loaded?'<span class="badge badge-ok">loaded</span>':'<span class="muted small">idle</span>'}</td>
+      <td style="padding:.4rem .25rem;text-align:right;white-space:nowrap">
+        ${loaded
+          ?`<button class="btn btn-ghost btn-sm" onclick="unloadModel(${idx})">Unload</button>`
+          :`<button class="btn btn-primary btn-sm" onclick="loadModel(${idx})">Load now</button>`}
+        <button class="btn btn-secondary btn-sm" onclick="openCfgModal(${idx})">Configure</button>
+        <button class="btn btn-ghost btn-sm" onclick="disableModel(${idx})">Remove</button>
+      </td>
+    </tr>`;
+  });
+  return '<div class="card" style="margin-top:1rem">'+
+    '<div class="card-title">Configured whisper-server models</div>'+
+    '<table style="width:100%;border-collapse:collapse;font-size:13px">'+
+    '<thead><tr style="color:var(--text-2);font-size:10px;text-transform:uppercase;letter-spacing:.05em">'+
+    '<th style="text-align:left;padding:.3rem .25rem;font-weight:700">Model</th>'+
+    '<th style="text-align:left;padding:.3rem .25rem;font-weight:700">Backend</th>'+
+    '<th style="text-align:left;padding:.3rem .25rem;font-weight:700">Model path</th>'+
+    '<th style="text-align:left;padding:.3rem .25rem;font-weight:700">Port / GPU</th>'+
+    '<th style="text-align:left;padding:.3rem .25rem;font-weight:700">Load mode</th>'+
+    '<th style="text-align:center;padding:.3rem .25rem;font-weight:700">Status</th>'+
+    '<th></th></tr></thead><tbody>'+rows.join('')+'</tbody></table></div>';
+}
+
 async function loadCachedModels(){
   _localModels = [];
   const hfEl   = document.getElementById('hf-models-list');
@@ -948,6 +1022,8 @@ async function loadCachedModels(){
     const r = await fetch('/admin/api/cached-models');
     if(!r.ok) throw new Error((await r.json()).detail||r.statusText);
     const d = await r.json();
+    const whisperModels = (await fetch('/admin/api/models').then(r=>r.ok?r.json():[]))
+      .filter(m => m.backend === 'whisper-server');
 
     // HF models
     const hf = d.hf||[];
@@ -1022,6 +1098,7 @@ async function loadCachedModels(){
         '<th style="text-align:center;padding:.3rem .25rem;font-weight:700">Config</th>'+
         '<th></th></tr></thead><tbody>'+rows.join('')+'</tbody></table>';
     }
+    ggufEl.insertAdjacentHTML('afterend', _renderWhisperServerRows(whisperModels));
   }catch(e){
     hfEl.innerHTML = ggufEl.innerHTML = `<span class="muted small">Error: ${esc(e.message)}</span>`;
   }
@@ -1103,6 +1180,9 @@ function openCfgModal(idx){
   document.getElementById('cfg-parser').value = s.parser || 'auto';
   document.getElementById('cfg-tools').checked = !!s.tools_closer_prompt;
   document.getElementById('cfg-grammar').checked = !!s.grammar_guided;
+  if (m.cacheType === 'whisper-server') {
+    document.getElementById('cfg-backend').value = 'cpu';
+  }
   openModal('cfg-modal');
 }
 
@@ -1150,6 +1230,31 @@ async function saveModelConfig(){
   }catch(e){ alert('Error: '+e.message); }
 }
 
+async function addWhisperServerModel(){
+  const usedVram = parseFloat(document.getElementById('ws-used-vram').value);
+  const payload = {
+    model_id: document.getElementById('ws-model-id').value.trim(),
+    model_type: 'audio_models',
+    backend: 'whisper-server',
+    server_path: document.getElementById('ws-server-path').value.trim(),
+    model_path: document.getElementById('ws-model-path').value.trim() || null,
+    port: parseInt(document.getElementById('ws-port').value, 10) || 8744,
+    gpu_device: parseInt(document.getElementById('ws-gpu-device').value, 10) || 0,
+    load_mode: document.getElementById('ws-load-mode').value,
+    used_vram_gb: Number.isNaN(usedVram) ? null : usedVram,
+  };
+  try{
+    const r = await fetch('/admin/api/model-configure', {
+      method:'POST',
+      headers:{'Content-Type':'application/json'},
+      body: JSON.stringify(payload)
+    });
+    const d = await r.json();
+    if(!r.ok) throw new Error(d.detail || 'Failed to add whisper-server model');
+    refreshLocal();
+  }catch(e){ alert('Error: '+e.message); }
+}
+
 async function loadModel(idx){
   const m = _localModels[idx];
   // Find the button and show loading state

codai/models/manager.py

@@ -1744,16 +1744,22 @@ class MultiModelManager:
             "embedding_models": "embedding",
         }
 
-        def _add(model_id: str, model_type: str = None):
+        def _add(model_id: str, model_type: str = None, meta: Dict[str, Any] = None):
             if model_id in seen_ids:
                 return
             seen_ids.add(model_id)
             caps = detect_model_capabilities(model_id)
             resolved_type = model_type or (caps.to_list()[0].split("_")[0] if caps.to_list() else "text")
+            meta = meta or {}
             models.append(ModelInfo(
                 id=model_id,
                 type=resolved_type,
                 capabilities=caps.to_list(),
+                backend=meta.get("backend"),
+                model_path=meta.get("model_path"),
+                port=meta.get("port"),
+                gpu_device=meta.get("gpu_device"),
+                load_mode=meta.get("load_mode"),
             ))
 
         # --- Models from config (the authoritative source) ---
@@ -1772,15 +1778,15 @@ class MultiModelManager:
                             mid = m.get("alias") or m.get("path") or m.get("id") or ""
                             raw = m.get("path") or m.get("id") or ""
                             if raw and raw != mid:
-                                _add(raw, mtype)
+                                _add(raw, mtype, m)
                                 short = raw.split("/")[-1] if "/" in raw else raw
                                 if short != raw:
-                                    _add(short, mtype)
+                                    _add(short, mtype, m)
                         if mid:
-                            _add(mid, mtype)
+                            _add(mid, mtype, m if isinstance(m, dict) else None)
                             short = mid.split("/")[-1] if "/" in mid else mid
                             if short != mid:
-                                _add(short, mtype)
+                                _add(short, mtype, m if isinstance(m, dict) else None)
         except Exception:
             pass
 

codai/pydantic/textrequest.py

@@ -121,8 +121,13 @@ class ModelInfo(BaseModel):
     owned_by: str = "huggingface"
     type: Optional[str] = None          # e.g. "text", "image", "video", "audio", "tts", "vision", "embedding"
     capabilities: Optional[List[str]] = None  # list of capability strings
+    backend: Optional[str] = None
+    model_path: Optional[str] = None
+    port: Optional[int] = None
+    gpu_device: Optional[int] = None
+    load_mode: Optional[str] = None
 
 
 class ModelList(BaseModel):
     object: str = "list"
-    data: List[ModelInfo]
+    data: List[ModelInfo]
\ No newline at end of file

tests/test_whisper_server_local_models.py

@@ -290,3 +290,44 @@ def test_settings_api_does_not_return_whisper_fields(monkeypatch):
     assert "whisper" not in response.json()
 
     app.dependency_overrides.clear()
+
+
+def test_models_template_contains_whisper_server_add_model_form():
+    template = Path("codai/admin/templates/models.html").read_text()
+
+    assert "Whisper-server simulated models" in template
+    assert "Add model" in template
+    assert "ws-model-id" in template
+    assert "ws-server-path" in template
+
+
+def test_settings_template_no_longer_contains_whisper_server_section():
+    template = Path("codai/admin/templates/settings.html").read_text()
+
+    assert "Whisper Server" not in template
+    assert "wsStart" not in template
+    assert "wsStop" not in template
+
+
+def test_model_info_supports_whisper_server_metadata_fields():
+    content = Path("codai/pydantic/textrequest.py").read_text()
+
+    assert "backend: Optional[str] = None" in content
+    assert "model_path: Optional[str] = None" in content
+    assert "port: Optional[int] = None" in content
+    assert "gpu_device: Optional[int] = None" in content
+    assert "load_mode: Optional[str] = None" in content
+
+
+def test_removed_whisper_server_admin_routes_return_not_found(monkeypatch):
+    from codai.admin import routes
+    from codai.api.app import app
+
+    app.dependency_overrides[routes.require_admin] = lambda: "admin"
+    client = TestClient(app)
+
+    assert client.get("/admin/api/whisper-server/status").status_code == 404
+    assert client.post("/admin/api/whisper-server/start", json={}).status_code == 404
+    assert client.post("/admin/api/whisper-server/stop", json={}).status_code == 404
+
+    app.dependency_overrides.clear()