omnius 1.0.169 → 1.0.170

package/dist/index.js

@@ -615859,9 +615859,10 @@ function buildRealtimeSystemPrompt(opts) {
       `- Produce one natural spoken turn, normally ${maxReplyWords} words or fewer.`,
       "- Use one sentence when possible; two short sentences only when repair or confirmation needs it.",
       "- Lead with the answer. Do not preface with status, analysis, summaries, or implementation narration.",
-      "- No markdown, bullets, tables, headings, citations, code blocks, JSON, or labels like 'Assistant:'.",
+      "- No markdown, bullets, tables, headings, citations, inline code, code blocks, JSON, or labels like 'Assistant:'.",
       "- Sound like a person on a live call: brief acknowledgment, direct answer, one focused follow-up only if needed.",
       "- If the ASR text is garbled or underspecified, ask a single compact repair question.",
+      "- Do not invent app modes, method names, settings, or implementation details when the caller has not supplied them.",
       "- Do not mention ASR, TTS, prompts, realtime mode, hidden reasoning, tools, or policy unless the caller explicitly asks.",
       "- If a request needs work outside this text-only exchange, say the next handoff in one short sentence."
     ].join("\n"),
@@ -615952,7 +615953,7 @@ function wordParts(text) {
 }
 function finalizeRealtimeReply(text, opts = {}) {
   const maxWords = clampInt2(opts.maxReplyWords, DEFAULT_REALTIME_MAX_REPLY_WORDS, 8, 80);
-  let clean5 = stripHiddenThinking(String(text ?? "")).replace(/```[\s\S]*?```/g, "").split("\n").map((line) => line.replace(/^\s*(?:[-*]+|\d+[.)])\s+/, "").trim()).filter(Boolean).join(" ").replace(/^(?:assistant|omnius|agent)\s*:\s*/i, "").replace(/\s+/g, " ").trim();
+  let clean5 = stripHiddenThinking(String(text ?? "")).replace(/```[\s\S]*?```/g, "").split("\n").map((line) => line.replace(/^\s*(?:[-*]+|\d+[.)])\s+/, "").trim()).filter(Boolean).join(" ").replace(/^(?:assistant|omnius|agent)\s*:\s*/i, "").replace(/`([^`]+)`/g, "$1").replace(/\s+/g, " ").trim();
   if (!clean5) return "I didn't catch that. Can you say it again?";
   const sentences = clean5.match(/[^.!?]+[.!?]+(?=\s|$)|[^.!?]+$/g) ?? [clean5];
   const selected = [];
@@ -657279,6 +657280,49 @@ function bodyString(body, keys) {
   }
   return "";
 }
+function realtimeFallbackCacheKey(ollamaUrl, missingModel) {
+  return `${ollamaUrl}
+${missingModel}`;
+}
+function isOllamaMissingModelError(body) {
+  return /model ['\"]?[^'\"]+['\"]? not found/i.test(body);
+}
+async function resolveRealtimeOllamaFallbackModel(ollamaUrl, timeoutMs, missingModel) {
+  try {
+    const cacheKey = realtimeFallbackCacheKey(ollamaUrl, missingModel);
+    const cached = realtimeOllamaFallbackCache.get(cacheKey);
+    if (cached) return cached;
+    const result = await ollamaRequest(ollamaUrl, "/api/tags", "GET", void 0, Math.min(timeoutMs, 1e4));
+    if (result.status >= 400) return null;
+    const parsed = JSON.parse(result.body);
+    const names = (parsed.models ?? []).map((entry) => typeof entry.name === "string" ? entry.name : typeof entry.model === "string" ? entry.model : "").filter(Boolean);
+    if (!names.length) return null;
+    const remember = (name10) => {
+      realtimeOllamaFallbackCache.set(cacheKey, name10);
+      return name10;
+    };
+    const exactLatest = `${missingModel}:latest`;
+    if (names.includes(exactLatest)) return remember(exactLatest);
+    const preferred = [
+      "qwen3.5-9b-r10:q4km",
+      "open-agents-qwen35-9b-r10-q4km:latest",
+      "open-agents-qwen35-9b-r10-parsed-q4km:latest",
+      "open-agents-qwen35-9b-r9-q4km:latest",
+      "qwen3:8b",
+      "open-agents-qwen3-8b:latest",
+      "omnius-qwen36-35b:latest",
+      "open-agents-qwen36:latest",
+      "qwen3.6:35b"
+    ];
+    for (const name10 of preferred) {
+      if (names.includes(name10)) return remember(name10);
+    }
+    const fallback = names.find((name10) => /qwen/i.test(name10) && !/embed|vision/i.test(name10)) ?? names.find((name10) => !/embed|vision|moondream/i.test(name10)) ?? null;
+    return fallback ? remember(fallback) : null;
+  } catch {
+    return null;
+  }
+}
 function realtimeEndpointMessages(body) {
   const messages2 = [];
   const suppliedSoul = bodyString(body, ["soul_md", "soul", "soulMd"]);
@@ -657305,13 +657349,14 @@ ${suppliedContext}` });
 }
 async function completeRealtimeTextOnly(opts) {
   const cfg = loadConfig();
-  const model = bodyString(opts.body, ["model"]) || cfg.model;
+  const requestedModel = bodyString(opts.body, ["model"]);
+  const model = requestedModel || opts.defaultModel || cfg.model;
   const route = resolveModelEndpoint(model);
   const limitErr = route?.endpoint ? checkEndpointRateLimit(route.endpoint) : null;
   if (limitErr) throw new Error(limitErr);
   const targetUrl = route?.endpoint.url ?? opts.ollamaUrl;
-  const targetType = route?.endpoint.type ?? cfg.backendType ?? "ollama";
-  const originalModel = route?.originalId ?? model.replace(/^[a-z]+\//, "");
+  const targetType = route?.endpoint.type ?? opts.defaultBackendType ?? cfg.backendType ?? "ollama";
+  let originalModel = route?.originalId ?? model.replace(/^[a-z]+\//, "");
   const realtimeOpts = {
     ...realtimeOptionsFromBody(opts.body, process.cwd(), opts.sessionId),
     surface: "voice_adapter"
@@ -657333,13 +657378,24 @@ async function completeRealtimeTextOnly(opts) {
   }
   const maxTokens = typeof requestBody["max_tokens"] === "number" ? requestBody["max_tokens"] : 120;
   const temperature = typeof requestBody["temperature"] === "number" ? requestBody["temperature"] : 0.6;
-  const result = await ollamaRequest(targetUrl, "/api/chat", "POST", JSON.stringify({
-    model: originalModel,
+  if (!requestedModel) {
+    originalModel = realtimeOllamaFallbackCache.get(realtimeFallbackCacheKey(targetUrl, originalModel)) ?? originalModel;
+  }
+  const makeOllamaChatBody = (modelName) => JSON.stringify({
+    model: modelName,
     messages: requestBody["messages"],
     stream: false,
     think: false,
     options: { temperature, num_predict: maxTokens }
-  }), timeoutMs, route?.endpoint);
+  });
+  let result = await ollamaRequest(targetUrl, "/api/chat", "POST", makeOllamaChatBody(originalModel), timeoutMs, route?.endpoint);
+  if (result.status >= 400 && !requestedModel && isOllamaMissingModelError(result.body)) {
+    const fallbackModel = await resolveRealtimeOllamaFallbackModel(targetUrl, timeoutMs, originalModel);
+    if (fallbackModel && fallbackModel !== originalModel) {
+      originalModel = fallbackModel;
+      result = await ollamaRequest(targetUrl, "/api/chat", "POST", makeOllamaChatBody(originalModel), timeoutMs, route?.endpoint);
+    }
+  }
   if (result.status >= 400) throw new Error(`Backend HTTP ${result.status}: ${result.body.slice(0, 300)}`);
   const parsed = JSON.parse(result.body);
   const rawReply = String(parsed?.message?.content ?? "").trim();
@@ -657354,7 +657410,7 @@ async function completeRealtimeTextOnly(opts) {
     }
   };
 }
-async function handleRealtimeText(req2, res, ollamaUrl) {
+async function handleRealtimeText(req2, res, ollamaUrl, defaults3 = {}) {
   const body = await parseJsonBody(req2);
   if (!body || typeof body !== "object") {
     jsonResponse(res, 400, { error: "invalid_request", message: "Expected a JSON object." });
@@ -657367,7 +657423,14 @@ async function handleRealtimeText(req2, res, ollamaUrl) {
   }
   try {
     const sessionId = typeof body["session_id"] === "string" ? body["session_id"] : void 0;
-    const result = await completeRealtimeTextOnly({ body, messages: messages2, ollamaUrl, sessionId });
+    const result = await completeRealtimeTextOnly({
+      body,
+      messages: messages2,
+      ollamaUrl,
+      defaultModel: defaults3.model,
+      defaultBackendType: defaults3.backendType,
+      sessionId
+    });
     const wantsPlain = String(req2.headers["accept"] ?? "").includes("text/plain") || body["format"] === "text";
     if (wantsPlain) {
       res.writeHead(200, { "Content-Type": "text/plain; charset=utf-8", "Cache-Control": "no-store" });
@@ -660331,7 +660394,7 @@ async function handlePostCommand(res, cmd) {
     });
   }
 }
-async function handleRequest(req2, res, ollamaUrl, verbose) {
+async function handleRequest(req2, res, ollamaUrl, verbose, runtimeDefaults = {}) {
   try {
     const _liveCfg = loadConfig();
     if (_liveCfg.backendUrl) ollamaUrl = _liveCfg.backendUrl;
@@ -660606,7 +660669,7 @@ async function handleRequest(req2, res, ollamaUrl, verbose) {
         status = 401;
         return;
       }
-      await handleRealtimeText(req2, res, ollamaUrl);
+      await handleRealtimeText(req2, res, ollamaUrl, runtimeDefaults);
       return;
     }
     if (pathname === "/v1/files" && method === "GET") {
@@ -662614,13 +662677,14 @@ ${historyLines}
       }));
     }
   } finally {
-    recordMetric(method, pathname, status);
+    const finalStatus = res.headersSent ? res.statusCode : status;
+    recordMetric(method, pathname, finalStatus);
     const latencyMs = Math.round(performance.now() - startMs);
     logRequest({
       requestId,
       method,
       path: pathname,
-      status,
+      status: finalStatus,
       latencyMs,
       user: req2._authUser ?? "anonymous",
       scope: req2._authScope ?? "none"
@@ -662630,7 +662694,7 @@ ${historyLines}
       requestId,
       method,
       path: pathname,
-      status,
+      status: finalStatus,
       user: req2._authUser ?? "anonymous",
       scope: req2._authScope ?? "none",
       latencyMs: Math.round(performance.now() - startMs),
@@ -663552,7 +663616,10 @@ function startApiServer(options2 = {}) {
       }
     } catch {
     }
-    handleRequest(req2, res, ollamaUrl, verbose).catch((err) => {
+    handleRequest(req2, res, ollamaUrl, verbose, {
+      model: options2.model ?? config.model,
+      backendType: options2.backendType ?? config.backendType
+    }).catch((err) => {
       metrics.totalErrors++;
       try {
         jsonResponse(res, 500, {
@@ -664374,7 +664441,9 @@ async function apiServeCommand(opts, config) {
     port: opts.port,
     // Let startApiServer() parse OMNIUS_HOST env if no explicit --port
     verbose: opts.verbose,
-    ollamaUrl: config.backendUrl
+    ollamaUrl: config.backendUrl,
+    model: config.model,
+    backendType: config.backendType
   });
   await new Promise((resolve57) => {
     server2.on("close", resolve57);
@@ -664429,7 +664498,7 @@ function setTimerEnabled(name10, enabled2) {
     return false;
   }
 }
-var require4, NEXUS_DIRECTORY_ORIGIN2, NEXUS_SPONSORS_URL2, endpointRegistry, modelRouteMap, endpointUsage, _lastEndpointDiagnostics, BACKEND_TIMEOUT_DEFAULT_MS, BACKEND_TIMEOUT_MAX_MS, MODEL_LIST_TIMEOUT_DEFAULT_MS, metrics, startedAt, runningProcesses, perKeyUsage, CRON_MARKER2;
+var require4, NEXUS_DIRECTORY_ORIGIN2, NEXUS_SPONSORS_URL2, endpointRegistry, modelRouteMap, endpointUsage, _lastEndpointDiagnostics, BACKEND_TIMEOUT_DEFAULT_MS, BACKEND_TIMEOUT_MAX_MS, MODEL_LIST_TIMEOUT_DEFAULT_MS, metrics, startedAt, realtimeOllamaFallbackCache, runningProcesses, perKeyUsage, CRON_MARKER2;
 var init_serve = __esm({
   "packages/cli/src/api/serve.ts"() {
     "use strict";
@@ -664477,6 +664546,7 @@ var init_serve = __esm({
       totalErrors: 0
     };
     startedAt = Date.now();
+    realtimeOllamaFallbackCache = /* @__PURE__ */ new Map();
     runningProcesses = /* @__PURE__ */ new Map();
     perKeyUsage = /* @__PURE__ */ new Map();
     CRON_MARKER2 = "# OMNIUS-SCHEDULED:";

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.169",
+  "version": "1.0.170",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.169",
+      "version": "1.0.170",
       "bundleDependencies": [
         "image-to-ascii"
       ],

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.169",
+  "version": "1.0.170",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",