bosun 0.36.4 → 0.36.5

package/monitor.mjs

@@ -991,7 +991,6 @@ let {
   fleet: fleetConfig,
   internalExecutor: internalExecutorConfig,
   executorMode: configExecutorMode,
-  githubReconcile: githubReconcileConfig,
 } = config;
 
 const telegramWeeklyReportEnabled = parseEnvBoolean(
@@ -1023,12 +1022,6 @@ let triggerSystemConfig =
     : { enabled: false, templates: [], defaults: { executor: "auto", model: "auto" } };
 let kanbanBackend = String(kanbanConfig?.backend || "internal").toLowerCase();
 let executorMode = configExecutorMode || getExecutorMode();
-let githubReconcile = githubReconcileConfig || {
-  enabled: false,
-  intervalMs: 5 * 60 * 1000,
-  mergedLookbackHours: 72,
-  trackingLabels: ["tracking"],
-};
 let chdirUnsupportedInRuntime = false;
 
 function isChdirUnsupportedError(err) {
@@ -10158,6 +10151,10 @@ function startTaskPlannerStatusLoop() {
   }, 25_000);
 }
 
+// GitHub reconciler hooks are currently optional; keep shutdown/reload calls safe.
+function restartGitHubReconciler() {}
+function stopGitHubReconciler() {}
+
 async function maybeTriggerTaskPlanner(reason, details, options = {}) {
   if (internalTaskExecutor?.isPaused?.()) {
     console.log("[monitor] task planner skipped: executor paused");
@@ -13056,6 +13053,61 @@ function formatOrchestratorTailForMonitorPrompt({
   }
 }
 
+const MONITOR_MONITOR_ACTIONABLE_DIGEST_MAX_AGE_MS = (() => {
+  const raw = Number(
+    process.env.DEVMODE_MONITOR_MONITOR_ACTIONABLE_DIGEST_MAX_AGE_MS ||
+      15 * 60 * 1000,
+  );
+  if (!Number.isFinite(raw) || raw <= 0) return 15 * 60 * 1000;
+  return Math.max(60 * 1000, Math.min(24 * 60 * 60 * 1000, Math.trunc(raw)));
+})();
+
+function parseDigestEntryTimestampMs(entry, { nowMs = Date.now(), digestStartedAt = 0 } = {}) {
+  const explicitTs = Number(entry?.timestamp ?? entry?.timeMs ?? entry?.ts);
+  if (Number.isFinite(explicitTs) && explicitTs > 0) return explicitTs;
+
+  const timeText = String(entry?.time || "").trim();
+  const match = timeText.match(/^(\d{2}):(\d{2}):(\d{2})$/);
+  if (!match) return null;
+
+  const hour = Number(match[1]);
+  const minute = Number(match[2]);
+  const second = Number(match[3]);
+  if (![hour, minute, second].every(Number.isFinite)) return null;
+
+  const nowDate = new Date(nowMs);
+  let candidate = Date.UTC(
+    nowDate.getUTCFullYear(),
+    nowDate.getUTCMonth(),
+    nowDate.getUTCDate(),
+    hour,
+    minute,
+    second,
+    0,
+  );
+
+  // Digest times are rendered as UTC HH:MM:SS.
+  if (candidate > nowMs + 60_000) {
+    candidate -= 24 * 60 * 60 * 1000;
+  }
+  // If digest started before midnight and entry time is after midnight.
+  if (
+    digestStartedAt > 0 &&
+    candidate < digestStartedAt - 60_000 &&
+    candidate + 24 * 60 * 60 * 1000 <= nowMs + 60_000
+  ) {
+    candidate += 24 * 60 * 60 * 1000;
+  }
+  return candidate;
+}
+
+function isDigestEntryActionable(entry, { nowMs = Date.now(), digestStartedAt = 0 } = {}) {
+  if (Number(entry?.priority || 99) > 3) return false;
+  const timestampMs = parseDigestEntryTimestampMs(entry, { nowMs, digestStartedAt });
+  if (!Number.isFinite(timestampMs)) return true;
+  return nowMs - timestampMs <= MONITOR_MONITOR_ACTIONABLE_DIGEST_MAX_AGE_MS;
+}
+
 async function buildMonitorMonitorPrompt({ trigger, entries, text }) {
   const digestSnapshot = getDigestSnapshot();
   const digestEntries =
@@ -13063,8 +13115,10 @@ async function buildMonitorMonitorPrompt({ trigger, entries, text }) {
       ? entries
       : digestSnapshot?.entries || [];
   const latestDigestText = String(text || monitorMonitor.lastDigestText || "");
-  const actionableEntries = digestEntries.filter(
-    (entry) => Number(entry?.priority || 99) <= 3,
+  const nowMs = Date.now();
+  const digestStartedAt = Number(digestSnapshot?.startedAt || 0);
+  const actionableEntries = digestEntries.filter((entry) =>
+    isDigestEntryActionable(entry, { nowMs, digestStartedAt }),
   );
   const modeHint =
     actionableEntries.length > 0 ? "reliability-fix" : "code-analysis";
@@ -14578,7 +14632,6 @@ function applyConfig(nextConfig, options = {}) {
   if (workflowAutomationEnabled) {
     ensureWorkflowAutomationEngine().catch(() => {});
   }
-  githubReconcile = nextConfig.githubReconcile || githubReconcile;
   agentPrompts = nextConfig.agentPrompts;
   configExecutorConfig = nextConfig.executorConfig;
   executorScheduler = nextConfig.scheduler;
@@ -14698,7 +14751,6 @@ function applyConfig(nextConfig, options = {}) {
   } else {
     stopMonitorMonitorSupervisor();
   }
-  restartGitHubReconciler();
 
   const nextArgs = scriptArgs?.join(" ") || "";
   const scriptChanged = prevScriptPath !== scriptPath || prevArgs !== nextArgs;
@@ -14732,7 +14784,6 @@ process.on("SIGINT", async () => {
   shuttingDown = true;
   stopWorkspaceSyncTimers();
   stopTaskPlannerStatusLoop();
-  stopGitHubReconciler();
   // Stop monitor-monitor immediately (it's safely restartable)
   stopMonitorMonitorSupervisor();
   if (vkLogStream) {
@@ -14786,7 +14837,6 @@ process.on("exit", () => {
   shuttingDown = true;
   stopWorkspaceSyncTimers();
   stopTaskPlannerStatusLoop();
-  stopGitHubReconciler();
   stopMonitorMonitorSupervisor();
   stopAgentAlertTailer();
   stopAgentWorkAnalyzer();
@@ -14802,7 +14852,6 @@ process.on("SIGTERM", async () => {
   shuttingDown = true;
   stopWorkspaceSyncTimers();
   stopTaskPlannerStatusLoop();
-  stopGitHubReconciler();
   // Stop monitor-monitor immediately (it's safely restartable)
   stopMonitorMonitorSupervisor();
   if (vkLogStream) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bosun",
-  "version": "0.36.4",
+  "version": "0.36.5",
   "description": "AI-powered orchestrator supervisor — manages AI agent executors with failover, auto-restarts on failure, analyzes crashes with Codex SDK, creates PRs via Vibe-Kanban API, and sends Telegram notifications. Supports N executors with weighted distribution, multi-repo projects, and auto-setup.",
   "type": "module",
   "license": "Apache 2.0",

package/setup-web-server.mjs

@@ -436,8 +436,8 @@ function buildStableSetupDefaults({
     workflowMaxConcurrentBranches: 8,
     voiceEnabled: true,
     voiceProvider: "auto",
-    voiceModel: "gpt-realtime-1.5",
-    voiceVisionModel: "gpt-4.1-mini",
+    voiceModel: "gpt-audio-1.5",
+    voiceVisionModel: "gpt-4.1-nano",
     voiceId: "alloy",
     voiceTurnDetection: "server_vad",
     voiceFallbackMode: "browser",
@@ -445,7 +445,7 @@ function buildStableSetupDefaults({
     openaiRealtimeApiKey: "",
     azureOpenaiRealtimeEndpoint: "",
     azureOpenaiRealtimeApiKey: "",
-    azureOpenaiRealtimeDeployment: "gpt-realtime-1.5",
+    azureOpenaiRealtimeDeployment: "gpt-audio-1.5",
     copilotEnableAllMcpTools: false,
     // Backward-compatible fields consumed by older setup UI revisions.
     distribution: "primary-only",
@@ -864,16 +864,16 @@ function applyNonBlockingSetupEnvDefaults(envMap, env = {}, sourceEnv = process.
       env.VOICE_MODEL,
       envMap.VOICE_MODEL,
       sourceEnv.VOICE_MODEL,
-    ) || "gpt-realtime-1.5",
-  ).trim() || "gpt-realtime-1.5";
+    ) || "gpt-audio-1.5",
+  ).trim() || "gpt-audio-1.5";
   envMap.VOICE_VISION_MODEL = String(
     pickNonEmptyValue(
       env.voiceVisionModel,
       env.VOICE_VISION_MODEL,
       envMap.VOICE_VISION_MODEL,
       sourceEnv.VOICE_VISION_MODEL,
-    ) || "gpt-4.1-mini",
-  ).trim() || "gpt-4.1-mini";
+    ) || "gpt-4.1-nano",
+  ).trim() || "gpt-4.1-nano";
   envMap.VOICE_ID = normalizeEnumValue(
     pickNonEmptyValue(
       env.voiceId,
@@ -953,8 +953,8 @@ function applyNonBlockingSetupEnvDefaults(envMap, env = {}, sourceEnv = process.
       env.AZURE_OPENAI_REALTIME_DEPLOYMENT,
       envMap.AZURE_OPENAI_REALTIME_DEPLOYMENT,
       sourceEnv.AZURE_OPENAI_REALTIME_DEPLOYMENT,
-    ) || "gpt-realtime-1.5",
-  ).trim() || "gpt-realtime-1.5";
+    ) || "gpt-audio-1.5",
+  ).trim() || "gpt-audio-1.5";
 
   envMap.CONTAINER_ENABLED = toBooleanEnvString(
     pickNonEmptyValue(env.containerEnabled, envMap.CONTAINER_ENABLED, sourceEnv.CONTAINER_ENABLED),

package/ui/modules/settings-schema.js

@@ -125,12 +125,12 @@ export const SETTINGS_SCHEMA = [
   // ── Voice Assistant ──────────────────────────────────────────
   { key: "VOICE_ENABLED",                  label: "Enable Voice Mode",          category: "voice", type: "boolean", defaultVal: true, description: "Enable the real-time voice assistant in the chat UI." },
   { key: "VOICE_PROVIDER",                 label: "Voice Provider",             category: "voice", type: "select", defaultVal: "auto", options: ["auto", "openai", "azure", "claude", "gemini", "fallback"], description: "Voice API provider. 'auto' selects based on available keys. 'fallback' uses browser speech APIs." },
-  { key: "VOICE_MODEL",                    label: "Voice Model",                category: "voice", type: "string", defaultVal: "gpt-4o-realtime-preview-2024-12-17", description: "OpenAI Realtime model to use for voice sessions." },
-  { key: "VOICE_VISION_MODEL",             label: "Vision Model",               category: "voice", type: "string", defaultVal: "gpt-4.1-mini", description: "Vision model used for live screen/camera understanding in voice mode." },
+  { key: "VOICE_MODEL",                    label: "Voice Model",                category: "voice", type: "select", defaultVal: "gpt-audio-1.5", options: ["gpt-audio-1.5", "gpt-realtime-1.5", "gpt-4o-realtime-preview-2024-12-17", "gemini-2.5-pro", "gemini-2.5-flash", "claude-3-7-sonnet-latest", "custom"], description: "Audio model for voice sessions. Select 'custom' to enter a model slug manually." },
+  { key: "VOICE_VISION_MODEL",             label: "Vision Model",               category: "voice", type: "select", defaultVal: "gpt-4.1-nano", options: ["gpt-4.1-nano", "gpt-4.1-mini", "gpt-4.1", "gemini-2.5-flash", "gemini-2.5-pro", "claude-3-7-sonnet-latest", "custom"], description: "Vision model for live screen/camera understanding. Select 'custom' to enter a model slug manually." },
   { key: "OPENAI_REALTIME_API_KEY",        label: "OpenAI Realtime Key",        category: "voice", type: "secret", sensitive: true, description: "Dedicated API key for voice. Falls back to OPENAI_API_KEY if not set." },
   { key: "AZURE_OPENAI_REALTIME_ENDPOINT", label: "Azure Realtime Endpoint",    category: "voice", type: "string", description: "Azure OpenAI endpoint for Realtime API (e.g., https://myresource.openai.azure.com).", validate: "^$|^https?://" },
   { key: "AZURE_OPENAI_REALTIME_API_KEY",  label: "Azure Realtime Key",         category: "voice", type: "secret", sensitive: true, description: "Azure OpenAI API key for Realtime API. Falls back to AZURE_OPENAI_API_KEY if not set." },
-  { key: "AZURE_OPENAI_REALTIME_DEPLOYMENT", label: "Azure Deployment",         category: "voice", type: "string", defaultVal: "gpt-4o-realtime-preview", description: "Azure deployment name for the Realtime model." },
+  { key: "AZURE_OPENAI_REALTIME_DEPLOYMENT", label: "Azure Deployment",         category: "voice", type: "select", defaultVal: "gpt-audio-1.5", options: ["gpt-audio-1.5", "gpt-realtime-1.5", "gpt-4o-realtime-preview", "custom"], description: "Azure deployment name for the Realtime model. Select 'custom' to enter manually." },
   { key: "VOICE_ID",                       label: "Voice",                      category: "voice", type: "select", defaultVal: "alloy", options: ["alloy", "ash", "ballad", "coral", "echo", "fable", "onyx", "nova", "sage", "shimmer", "verse"], description: "Voice personality for text-to-speech output." },
   { key: "VOICE_TURN_DETECTION",           label: "Turn Detection",             category: "voice", type: "select", defaultVal: "server_vad", options: ["server_vad", "semantic_vad", "none"], description: "How the model detects when you stop speaking. 'semantic_vad' is more intelligent but higher latency." },
   { key: "VOICE_DELEGATE_EXECUTOR",        label: "Delegate Executor",          category: "voice", type: "select", defaultVal: "codex-sdk", options: ["codex-sdk", "copilot-sdk", "claude-sdk", "gemini-sdk", "opencode-sdk"], description: "Which agent executor voice tool calls delegate to for complex tasks." },
@@ -307,8 +307,11 @@ export function validateSetting(def, value) {
         return { valid: false, error: "Must be true or false" };
       return { valid: true };
     case "select":
-      if (def.options && !def.options.includes(String(value)))
+      if (def.options && !def.options.includes(String(value))) {
+        // Allow arbitrary values when the schema includes "custom" as an option
+        if (def.options.includes("custom")) return { valid: true };
         return { valid: false, error: `Must be one of: ${def.options.join(", ")}` };
+      }
       return { valid: true };
     default:
       if (def.validate) {

package/ui/setup.html

@@ -705,8 +705,8 @@ function App() {
   // Voice assistant
   const [voiceEnabled, setVoiceEnabled] = useState(true);
   const [voiceProvider, setVoiceProvider] = useState("auto");
-  const [voiceModel, setVoiceModel] = useState("gpt-4o-realtime-preview-2024-12-17");
-  const [voiceVisionModel, setVoiceVisionModel] = useState("gpt-4.1-mini");
+  const [voiceModel, setVoiceModel] = useState("gpt-audio-1.5");
+  const [voiceVisionModel, setVoiceVisionModel] = useState("gpt-4.1-nano");
   const [voiceId, setVoiceId] = useState("alloy");
   const [voiceTurnDetection, setVoiceTurnDetection] = useState("server_vad");
   const [voiceFallbackMode, setVoiceFallbackMode] = useState("browser");
@@ -714,7 +714,7 @@ function App() {
   const [openaiRealtimeApiKey, setOpenaiRealtimeApiKey] = useState("");
   const [azureOpenaiRealtimeEndpoint, setAzureOpenaiRealtimeEndpoint] = useState("");
   const [azureOpenaiRealtimeApiKey, setAzureOpenaiRealtimeApiKey] = useState("");
-  const [azureOpenaiRealtimeDeployment, setAzureOpenaiRealtimeDeployment] = useState("gpt-4o-realtime-preview");
+  const [azureOpenaiRealtimeDeployment, setAzureOpenaiRealtimeDeployment] = useState("gpt-audio-1.5");
   const [voiceProviders, setVoiceProviders] = useState([
     {
       id: Date.now(),
@@ -727,11 +727,11 @@ function App() {
   ]);
 
   const VOICE_PROVIDER_MODEL_DEFAULTS = {
-    openai: { model: "gpt-4o-realtime-preview-2024-12-17", visionModel: "gpt-4.1-mini" },
-    azure: { model: "gpt-4o-realtime-preview-2024-12-17", visionModel: "gpt-4.1-mini" },
-    claude: { model: "claude-3-7-sonnet-latest", visionModel: "claude-3-7-sonnet-latest" },
-    gemini: { model: "gemini-2.5-pro", visionModel: "gemini-2.5-flash" },
-    fallback: { model: "", visionModel: "" },
+    openai: { model: "gpt-audio-1.5", visionModel: "gpt-4.1-nano", models: ["gpt-audio-1.5", "gpt-realtime-1.5", "gpt-4o-realtime-preview-2024-12-17"], visionModels: ["gpt-4.1-nano", "gpt-4.1-mini", "gpt-4.1"] },
+    azure: { model: "gpt-audio-1.5", visionModel: "gpt-4.1-nano", models: ["gpt-audio-1.5", "gpt-realtime-1.5", "gpt-4o-realtime-preview"], visionModels: ["gpt-4.1-nano", "gpt-4.1-mini", "gpt-4.1"] },
+    claude: { model: "claude-3-7-sonnet-latest", visionModel: "claude-3-7-sonnet-latest", models: ["claude-3-7-sonnet-latest", "claude-sonnet-4-20250514"], visionModels: ["claude-3-7-sonnet-latest", "claude-sonnet-4-20250514"] },
+    gemini: { model: "gemini-2.5-pro", visionModel: "gemini-2.5-flash", models: ["gemini-2.5-pro", "gemini-2.5-flash"], visionModels: ["gemini-2.5-flash", "gemini-2.5-pro"] },
+    fallback: { model: "", visionModel: "", models: [], visionModels: [] },
   };
 
   const getVoiceProviderModelDefaults = (provider) =>
@@ -748,7 +748,7 @@ function App() {
     const normalizedAzureDeployment = String(
       entry.azureDeployment ??
       fallback.azureDeployment ??
-      (normalizedProvider === "azure" ? "gpt-4o-realtime-preview" : ""),
+      (normalizedProvider === "azure" ? "gpt-audio-1.5" : ""),
     ).trim();
     return {
       id: entry.id || Date.now() + Math.random(),
@@ -783,7 +783,7 @@ function App() {
           visionModel: defaults_.visionModel,
           azureDeployment:
             nextProvider === "azure"
-              ? String(target.azureDeployment || "gpt-4o-realtime-preview")
+              ? String(target.azureDeployment || "gpt-audio-1.5")
               : "",
         });
         return list;
@@ -1157,10 +1157,10 @@ function App() {
       if (env.AZURE_OPENAI_REALTIME_API_KEY) { setAzureOpenaiRealtimeApiKey(env.AZURE_OPENAI_REALTIME_API_KEY); envLoaded = true; }
       if (env.AZURE_OPENAI_REALTIME_DEPLOYMENT) { setAzureOpenaiRealtimeDeployment(env.AZURE_OPENAI_REALTIME_DEPLOYMENT); envLoaded = true; }
       const resolvedVoiceProvider = String(env.VOICE_PROVIDER || existingVoice.provider || d.voiceProvider || voiceProvider || "openai").trim().toLowerCase();
-      const resolvedVoiceModel = String(env.VOICE_MODEL || existingVoice.model || d.voiceModel || voiceModel || "gpt-4o-realtime-preview-2024-12-17").trim();
-      const resolvedVoiceVisionModel = String(env.VOICE_VISION_MODEL || existingVoice.visionModel || d.voiceVisionModel || voiceVisionModel || "gpt-4.1-mini").trim();
+      const resolvedVoiceModel = String(env.VOICE_MODEL || existingVoice.model || d.voiceModel || voiceModel || "gpt-audio-1.5").trim();
+      const resolvedVoiceVisionModel = String(env.VOICE_VISION_MODEL || existingVoice.visionModel || d.voiceVisionModel || voiceVisionModel || "gpt-4.1-nano").trim();
       const resolvedVoiceId = String(env.VOICE_ID || existingVoice.voiceId || d.voiceId || voiceId || "alloy").trim();
-      const resolvedAzureDeployment = String(env.AZURE_OPENAI_REALTIME_DEPLOYMENT || existingVoice.azureDeployment || d.azureOpenaiRealtimeDeployment || azureOpenaiRealtimeDeployment || "gpt-4o-realtime-preview").trim();
+      const resolvedAzureDeployment = String(env.AZURE_OPENAI_REALTIME_DEPLOYMENT || existingVoice.azureDeployment || d.azureOpenaiRealtimeDeployment || azureOpenaiRealtimeDeployment || "gpt-audio-1.5").trim();
       if (Array.isArray(existingVoice.providers) && existingVoice.providers.length > 0) {
         setVoiceProviders(
           normalizeVoiceProviders(existingVoice.providers, {
@@ -2521,23 +2521,25 @@ function App() {
 
   // ── Step 7: Advanced Settings ──────────────────────────────────────────────
 
-  const StepAdvanced = () => {
-    // Section is rendered as a real Preact component (`<${Section}/>`) so its
-    // useState call is properly isolated to its own fiber — safe to use here.
-    function Section({ id, title, defaultOpen, children }) {
-      const [open, setOpen] = useState(!!defaultOpen);
-      return html`
-        <div style="border:1px solid var(--border-primary);border-radius:var(--radius-sm);margin-bottom:10px;overflow:hidden">
-          <div style="display:flex;align-items:center;justify-content:space-between;padding:12px 16px;cursor:pointer;background:var(--bg-input);user-select:none"
-            onclick=${() => setOpen((v) => !v)}>
-            <span style="font-weight:600;font-size:0.88rem">${iconText(title)}</span>
-            <span style="color:var(--text-dim);font-size:0.8rem">${open ? "▲" : "▼"}</span>
-          </div>
-          ${open && html`<div style="padding:16px">${children}</div>`}
+  // Section is defined at module scope so its identity is stable across
+  // re-renders. Defining it inside StepAdvanced would create a new function
+  // reference on every state update, causing Preact to unmount/remount the
+  // component (losing open state + input focus).
+  function Section({ id, title, defaultOpen, children }) {
+    const [open, setOpen] = useState(!!defaultOpen);
+    return html`
+      <div style="border:1px solid var(--border-primary);border-radius:var(--radius-sm);margin-bottom:10px;overflow:hidden">
+        <div style="display:flex;align-items:center;justify-content:space-between;padding:12px 16px;cursor:pointer;background:var(--bg-input);user-select:none"
+          onclick=${() => setOpen((v) => !v)}>
+          <span style="font-weight:600;font-size:0.88rem">${iconText(title)}</span>
+          <span style="color:var(--text-dim);font-size:0.8rem">${open ? "▲" : "▼"}</span>
         </div>
-      `;
-    }
+        ${open && html`<div style="padding:16px">${children}</div>`}
+      </div>
+    `;
+  }
 
+  const StepAdvanced = () => {
     return html`
       <h2>Advanced Settings</h2>
       <p class="step-desc">Fine-tune execution, model profiles, and infrastructure. Defaults are sensible — only change what you need.</p>
@@ -2876,21 +2878,73 @@ function App() {
                 </div>
                 <div class="form-group">
                   <label>Model</label>
-                  <input
-                    type="text"
-                    value=${providerRow.model}
-                    oninput=${(e) => updateVoiceProviderRow(idx, "model", e.target.value)}
-                    placeholder="Provider model"
-                  />
+                  ${(() => {
+                    const defaults_ = getVoiceProviderModelDefaults(providerRow.provider);
+                    const knownModels = defaults_.models || [];
+                    const isCustom = knownModels.length > 0 && !knownModels.includes(providerRow.model) && providerRow.model !== "";
+                    return html`
+                      <select value=${isCustom ? "custom" : providerRow.model}
+                        onchange=${(e) => {
+                          if (e.target.value === "custom") {
+                            updateVoiceProviderRow(idx, "model", "");
+                          } else {
+                            updateVoiceProviderRow(idx, "model", e.target.value);
+                          }
+                        }}>
+                        ${knownModels.map((m) => html`<option value=${m}>${m}</option>`)}
+                        <option value="custom">custom...</option>
+                      </select>
+                      ${(isCustom || (knownModels.length > 0 && !knownModels.includes(providerRow.model))) && html`
+                        <input type="text" value=${providerRow.model}
+                          oninput=${(e) => updateVoiceProviderRow(idx, "model", e.target.value)}
+                          placeholder="Enter custom model slug..."
+                          style="margin-top:4px"
+                        />
+                      `}
+                      ${knownModels.length === 0 && html`
+                        <input type="text" value=${providerRow.model}
+                          oninput=${(e) => updateVoiceProviderRow(idx, "model", e.target.value)}
+                          placeholder="Provider model"
+                          style="margin-top:4px"
+                        />
+                      `}
+                    `;
+                  })()}
                 </div>
                 <div class="form-group">
                   <label>Vision Model</label>
-                  <input
-                    type="text"
-                    value=${providerRow.visionModel}
-                    oninput=${(e) => updateVoiceProviderRow(idx, "visionModel", e.target.value)}
-                    placeholder="Provider vision model"
-                  />
+                  ${(() => {
+                    const defaults_ = getVoiceProviderModelDefaults(providerRow.provider);
+                    const knownVisionModels = defaults_.visionModels || [];
+                    const isCustomVision = knownVisionModels.length > 0 && !knownVisionModels.includes(providerRow.visionModel) && providerRow.visionModel !== "";
+                    return html`
+                      <select value=${isCustomVision ? "custom" : providerRow.visionModel}
+                        onchange=${(e) => {
+                          if (e.target.value === "custom") {
+                            updateVoiceProviderRow(idx, "visionModel", "");
+                          } else {
+                            updateVoiceProviderRow(idx, "visionModel", e.target.value);
+                          }
+                        }}>
+                        ${knownVisionModels.map((m) => html`<option value=${m}>${m}</option>`)}
+                        <option value="custom">custom...</option>
+                      </select>
+                      ${(isCustomVision || (knownVisionModels.length > 0 && !knownVisionModels.includes(providerRow.visionModel))) && html`
+                        <input type="text" value=${providerRow.visionModel}
+                          oninput=${(e) => updateVoiceProviderRow(idx, "visionModel", e.target.value)}
+                          placeholder="Enter custom vision model slug..."
+                          style="margin-top:4px"
+                        />
+                      `}
+                      ${knownVisionModels.length === 0 && html`
+                        <input type="text" value=${providerRow.visionModel}
+                          oninput=${(e) => updateVoiceProviderRow(idx, "visionModel", e.target.value)}
+                          placeholder="Provider vision model"
+                          style="margin-top:4px"
+                        />
+                      `}
+                    `;
+                  })()}
                 </div>
                 <div class="form-group">
                   <label>Voice Persona</label>
@@ -2915,7 +2969,7 @@ function App() {
                       type="text"
                       value=${providerRow.azureDeployment || ""}
                       oninput=${(e) => updateVoiceProviderRow(idx, "azureDeployment", e.target.value)}
-                      placeholder="gpt-4o-realtime-preview"
+                      placeholder="gpt-audio-1.5"
                     />
                   </div>
                 `}

package/voice-agents-sdk.mjs

@@ -50,7 +50,7 @@ async function getGoogleGenAI() {
 
 // ── Constants ───────────────────────────────────────────────────────────────
 
-const OPENAI_REALTIME_MODEL = "gpt-realtime-1.5";
+const OPENAI_REALTIME_MODEL = "gpt-audio-1.5";
 const GEMINI_LIVE_MODEL = "gemini-2.5-flash-native-audio-preview-12-2025";
 
 const SDK_PROVIDERS = Object.freeze({
@@ -316,7 +316,7 @@ export async function connectRealtimeSession(sessionHandle, config = {}) {
     }
     const endpoint = String(config.azureEndpoint || "").trim().replace(/\/+$/, "");
     const deployment = String(
-      config.azureDeployment || "gpt-realtime-1.5",
+      config.azureDeployment || "gpt-audio-1.5",
     ).trim();
     connectOpts.apiKey = credential;
     connectOpts.url = `${endpoint}/openai/realtime?api-version=2025-04-01-preview&deployment=${deployment}`;

package/voice-relay.mjs

@@ -22,9 +22,9 @@ let _configLoadedAt = 0;   // timestamp of last config load
 const CONFIG_TTL_MS = 30_000; // re-read config every 30s
 
 const OPENAI_REALTIME_URL = "https://api.openai.com/v1/realtime";
-const OPENAI_REALTIME_MODEL = "gpt-realtime-1.5"; // Released 2026-02-23; replaces gpt-4o-realtime-preview
+const OPENAI_REALTIME_MODEL = "gpt-audio-1.5";
 const OPENAI_RESPONSES_URL = "https://api.openai.com/v1/responses";
-const OPENAI_DEFAULT_VISION_MODEL = "gpt-4.1-mini";
+const OPENAI_DEFAULT_VISION_MODEL = "gpt-4.1-nano";
 
 const AZURE_API_VERSION = "2025-04-01-preview";
 const ANTHROPIC_MESSAGES_URL = "https://api.anthropic.com/v1/messages";