omnius 1.0.209 → 1.0.211

package/dist/index.js

@@ -284705,6 +284705,7 @@ async function launchService() {
       SCRAPE_REQUIRE_AUTH: "0"
     }
   });
+  serviceProcess.unref();
   const cleanupService = () => {
     if (serviceProcess && serviceProcess.pid && !serviceProcess.killed) {
       try {
@@ -284958,6 +284959,7 @@ var init_browser_action = __esm({
               await apiCall("/session/close");
             } catch {
             }
+            killBrowserActionServicePort();
             activeSessionId = null;
             activeSessionHeadless = null;
             activeSessionUrl = null;
@@ -561827,6 +561829,30 @@ ${blob}
       _buildToolFingerprint(name10, args) {
         return `${name10}:${this._buildExactArgsKey(args)}`;
       }
+      _dedupeToolCallsForResponse(toolCalls, turn) {
+        if (toolCalls.length <= 1)
+          return toolCalls;
+        const seen = /* @__PURE__ */ new Set();
+        const deduped = [];
+        let dropped = 0;
+        for (const tc of toolCalls) {
+          const fp = this._buildToolFingerprint(tc.name, tc.arguments ?? {});
+          if (seen.has(fp)) {
+            dropped++;
+            continue;
+          }
+          seen.add(fp);
+          deduped.push(tc);
+        }
+        if (dropped > 0) {
+          this.emit({
+            type: "status",
+            content: `Response dedupe: dropped ${dropped} exact duplicate tool call(s) before execution (turn ${turn})`,
+            timestamp: (/* @__PURE__ */ new Date()).toISOString()
+          });
+        }
+        return deduped;
+      }
       _decodeToolFingerprint(fingerprint) {
         const colonIdx = fingerprint.indexOf(":");
         const toolName = colonIdx > 0 ? fingerprint.slice(0, colonIdx) : fingerprint;
@@ -564687,6 +564713,14 @@ ${memoryLines.join("\n")}`
           if (msg.toolCalls && msg.toolCalls.length > 0) {
             consecutiveTextOnly = 0;
             consecutiveThinkOnly = 0;
+            msg.toolCalls = this._dedupeToolCallsForResponse(msg.toolCalls, turn);
+            if (msg.toolCalls.length === 0) {
+              messages2.push({
+                role: "system",
+                content: "[RESPONSE DEDUPE] All tool calls in the last response were exact duplicates and were dropped. Use the prior results already in context/cache to continue."
+              });
+              continue;
+            }
             const _RESPONSE_CALL_CAPS = {
               small: 2,
               medium: 4,
@@ -567330,6 +567364,14 @@ ${this.options.maxTurns && this.options.maxTurns > 0 ? `You have ${this.options.
             if (msg.toolCalls && msg.toolCalls.length > 0) {
               consecutiveTextOnly = 0;
               consecutiveThinkOnly = 0;
+              msg.toolCalls = this._dedupeToolCallsForResponse(msg.toolCalls, turn);
+              if (msg.toolCalls.length === 0) {
+                messages2.push({
+                  role: "system",
+                  content: "[RESPONSE DEDUPE] All tool calls in the last response were exact duplicates and were dropped. Use the prior results already in context/cache to continue."
+                });
+                continue;
+              }
               messages2.push({
                 role: "assistant",
                 content: msg.content || null,
@@ -572270,6 +572312,54 @@ ${description}`
           releasePoolSlot(poolSuccess);
         }
       }
+      async nativeOllamaChatCompletion(request) {
+        const cleanedMessages = applyMemoryPrefixToMessages(normalizeMessagesForStrictOpenAI(sanitizeHistoryThink(request.messages)), request.memoryPrefix);
+        const requestMessages = injectNoThinkDirective(cleanedMessages);
+        const responseFormat = request.responseFormat ?? request.response_format;
+        const options2 = {
+          temperature: request.temperature,
+          num_predict: request.maxTokens
+        };
+        const reqNumCtx = request.numCtx ?? request.num_ctx;
+        if (Number.isFinite(reqNumCtx) && (reqNumCtx ?? 0) > 0) {
+          options2["num_ctx"] = reqNumCtx;
+        }
+        const body = {
+          model: this.model,
+          messages: requestMessages,
+          stream: false,
+          think: false,
+          options: options2
+        };
+        if (responseFormat !== void 0) {
+          body["format"] = "json";
+        }
+        const effectiveTimeoutMs = Number.isFinite(request.timeoutMs) && request.timeoutMs > 0 ? request.timeoutMs : 0;
+        const timeoutSignal = effectiveTimeoutMs > 0 && typeof AbortSignal.timeout === "function" ? AbortSignal.timeout(effectiveTimeoutMs) : void 0;
+        const fetchOpts = {
+          method: "POST",
+          headers: this.authHeaders(),
+          body: JSON.stringify(body)
+        };
+        if (timeoutSignal)
+          fetchOpts.signal = timeoutSignal;
+        const resp = await fetch(`${this.baseUrl}/api/chat`, fetchOpts);
+        if (!resp.ok) {
+          const text = await resp.text().catch(() => "");
+          throw new Error(`Ollama native HTTP ${resp.status}: ${backendHttpErrorDetail(text)}`);
+        }
+        const data = await resp.json();
+        const message2 = data["message"] ?? {};
+        const content = typeof message2["content"] === "string" ? message2["content"] : "";
+        return {
+          choices: [{ message: { content: content || null } }],
+          usage: buildAgenticUsage({
+            totalTokens: numberFromUnknown2(data["prompt_eval_count"]) ?? 0,
+            promptEvalCount: numberFromUnknown2(data["prompt_eval_count"]),
+            completionTokens: numberFromUnknown2(data["eval_count"])
+          })
+        };
+      }
       /** Anthropic Messages API translation — converts our standard format to/from Anthropic's. */
       async _anthropicChatCompletion(request) {
         const systemMsgs = request.messages.filter((m2) => m2.role === "system");
@@ -641965,7 +642055,7 @@ function renderTelegramSubAgentError(username, error) {
   process.stdout.write(`    ${c3.dim("│")} ${c3.magenta("✘")} @${username}: ${c3.dim(preview)}
 `);
 }
-var TELEGRAM_TOOL_ACTION_GROUPS, TELEGRAM_TOOL_ACTION_GROUP, TELEGRAM_TOOL_MUTATING_GROUPS, DEFAULT_TELEGRAM_TOOL_GROUP_POLICY, TELEGRAM_TOOL_BUTTON_LABELS, TELEGRAM_SAFETY_PROMPT, ADMIN_DM_PROMPT, ADMIN_GROUP_PROMPT, TELEGRAM_PUBLIC_SOUL_PROFILE, TELEGRAM_PUBLIC_ORCHESTRATOR_CONTRACT, TELEGRAM_PUBLIC_MEMORY_SCOPE_CONTRACT, TELEGRAM_PUBLIC_VISION_STACK_CONTRACT, GROUP_REPLY_DISCRETION_PROMPT, TELEGRAM_CHAT_MODE_PROMPT, ADMIN_CHAT_PROFILE_PROMPT, TELEGRAM_ACTION_RESPONSE_CONTRACT, TELEGRAM_EXTERNAL_ACQUISITION_CONTRACT, TELEGRAM_LINK_INTEGRITY_CONTRACT, TELEGRAM_INTERACTION_DECISION_RESPONSE_FORMAT, TELEGRAM_INTERACTION_DECISION_MINIMAL_SCHEMA, TELEGRAM_INTERACTION_DECISION_REPAIR_SCHEMA, TELEGRAM_CHAT_REPLY_RESPONSE_FORMAT, TELEGRAM_SPACED_URL_RE, TELEGRAM_HTTP_URL_RE, TELEGRAM_STUCK_SELF_TALK_PREFIXES, TELEGRAM_CHAT_HISTORY_LIMIT, TELEGRAM_CONTEXT_RECENT_DEFAULT, TELEGRAM_CONTEXT_LINE_LIMIT, TELEGRAM_CONTEXT_SAMPLE_LIMIT, TELEGRAM_MEMORY_CARD_LIMIT, TELEGRAM_MEMORY_NOTE_LIMIT, TELEGRAM_ASSOCIATIVE_FACT_LIMIT, TELEGRAM_ASSOCIATIVE_USER_FACT_LIMIT, TELEGRAM_ASSOCIATIVE_ACTION_LIMIT, TELEGRAM_ASSOCIATIVE_RELATION_LIMIT, TELEGRAM_MEMORY_STOPWORDS, TELEGRAM_MEMORY_GENERIC_QUERY_TOKENS, TELEGRAM_SUB_AGENT_BOUNDED_OPTIONS, TELEGRAM_SUB_AGENT_DEFAULT_LIMIT, TELEGRAM_SUB_AGENT_MAX_LIMIT, TELEGRAM_SUB_AGENT_BURST_CONTEXT_LIMIT, TELEGRAM_ADMIN_LIVE_PANEL_PAGES, TELEGRAM_ADMIN_LIVE_MUTATION_TOOLS, TELEGRAM_PUBLIC_HELP_COMMANDS2, TELEGRAM_REMINDER_SLASH_COMMANDS, TELEGRAM_REFLECTION_SLASH_COMMANDS, TELEGRAM_PUBLIC_BOT_COMMAND_NAMES, TELEGRAM_IMAGE_EXTENSIONS, MEDIA_CACHE_TTL_MS, TELEGRAM_CHANNEL_DMN_SWEEP_MS, TELEGRAM_CHANNEL_DMN_IDLE_AFTER_MS, TELEGRAM_CHANNEL_DMN_MIN_INTERVAL_MS, TELEGRAM_CHANNEL_DMN_MIN_MESSAGES, TELEGRAM_ALLOWED_UPDATES, TELEGRAM_DEFAULT_LONG_POLL_TIMEOUT_SECONDS, TELEGRAM_DEFAULT_ROUTER_MODEL_CANDIDATES, TELEGRAM_PUBLIC_TOOL_QUOTAS, TelegramBridge;
+var TELEGRAM_TOOL_ACTION_GROUPS, TELEGRAM_TOOL_ACTION_GROUP, TELEGRAM_TOOL_MUTATING_GROUPS, DEFAULT_TELEGRAM_TOOL_GROUP_POLICY, TELEGRAM_TOOL_BUTTON_LABELS, TELEGRAM_SAFETY_PROMPT, ADMIN_DM_PROMPT, ADMIN_GROUP_PROMPT, TELEGRAM_PUBLIC_SOUL_PROFILE, TELEGRAM_PUBLIC_ORCHESTRATOR_CONTRACT, TELEGRAM_PUBLIC_MEMORY_SCOPE_CONTRACT, TELEGRAM_PUBLIC_VISION_STACK_CONTRACT, GROUP_REPLY_DISCRETION_PROMPT, TELEGRAM_CHAT_MODE_PROMPT, ADMIN_CHAT_PROFILE_PROMPT, TELEGRAM_ACTION_RESPONSE_CONTRACT, TELEGRAM_EXTERNAL_ACQUISITION_CONTRACT, TELEGRAM_LINK_INTEGRITY_CONTRACT, TELEGRAM_INTERACTION_DECISION_RESPONSE_FORMAT, TELEGRAM_INTERACTION_DECISION_MINIMAL_SCHEMA, TELEGRAM_INTERACTION_DECISION_REPAIR_SCHEMA, TELEGRAM_CHAT_REPLY_RESPONSE_FORMAT, TELEGRAM_SPACED_URL_RE, TELEGRAM_HTTP_URL_RE, TELEGRAM_STUCK_SELF_TALK_PREFIXES, TELEGRAM_CHAT_HISTORY_LIMIT, TELEGRAM_CONTEXT_RECENT_DEFAULT, TELEGRAM_CONTEXT_LINE_LIMIT, TELEGRAM_CONTEXT_SAMPLE_LIMIT, TELEGRAM_MEMORY_CARD_LIMIT, TELEGRAM_MEMORY_NOTE_LIMIT, TELEGRAM_ASSOCIATIVE_FACT_LIMIT, TELEGRAM_ASSOCIATIVE_USER_FACT_LIMIT, TELEGRAM_ASSOCIATIVE_ACTION_LIMIT, TELEGRAM_ASSOCIATIVE_RELATION_LIMIT, TELEGRAM_MEMORY_STOPWORDS, TELEGRAM_MEMORY_GENERIC_QUERY_TOKENS, TELEGRAM_SUB_AGENT_BOUNDED_OPTIONS, TELEGRAM_SUB_AGENT_DEFAULT_LIMIT, TELEGRAM_SUB_AGENT_MAX_LIMIT, TELEGRAM_SUB_AGENT_BURST_CONTEXT_LIMIT, TELEGRAM_ADMIN_LIVE_PANEL_PAGES, TELEGRAM_ADMIN_LIVE_MUTATION_TOOLS, TELEGRAM_PUBLIC_HELP_COMMANDS2, TELEGRAM_REMINDER_SLASH_COMMANDS, TELEGRAM_REFLECTION_SLASH_COMMANDS, TELEGRAM_PUBLIC_BOT_COMMAND_NAMES, TELEGRAM_IMAGE_EXTENSIONS, MEDIA_CACHE_TTL_MS, TELEGRAM_CHANNEL_DMN_SWEEP_MS, TELEGRAM_CHANNEL_DMN_IDLE_AFTER_MS, TELEGRAM_CHANNEL_DMN_MIN_INTERVAL_MS, TELEGRAM_CHANNEL_DMN_MIN_MESSAGES, TELEGRAM_ALLOWED_UPDATES, TELEGRAM_DEFAULT_LONG_POLL_TIMEOUT_SECONDS, TELEGRAM_ROUTER_AUTO_MIN_PARAMETERS_B, TELEGRAM_PUBLIC_TOOL_QUOTAS, TelegramBridge;
 var init_telegram_bridge = __esm({
   "packages/cli/src/tui/telegram-bridge.ts"() {
     "use strict";
@@ -642428,20 +642518,7 @@ Telegram link integrity contract:
     TELEGRAM_CHANNEL_DMN_MIN_MESSAGES = 4;
     TELEGRAM_ALLOWED_UPDATES = ["message", "guest_message", "callback_query", "poll", "message_reaction", "message_reaction_count"];
     TELEGRAM_DEFAULT_LONG_POLL_TIMEOUT_SECONDS = 50;
-    TELEGRAM_DEFAULT_ROUTER_MODEL_CANDIDATES = [
-      "qwen2.5:3b",
-      "qwen2.5:7b",
-      "llama3.2:1b",
-      "llama3.2:3b",
-      "gemma3:1b",
-      "gemma3:4b",
-      "phi3:mini",
-      "phi4-mini:latest",
-      "qwen3:0.6b",
-      "qwen3:1.7b",
-      "qwen3:4b",
-      "qwen3:8b"
-    ];
+    TELEGRAM_ROUTER_AUTO_MIN_PARAMETERS_B = 8;
     TELEGRAM_PUBLIC_TOOL_QUOTAS = {
       web: { limit: 20, windowMs: 60 * 6e4 },
       media: { limit: 30, windowMs: 60 * 6e4 },
@@ -647277,8 +647354,13 @@ ${this.quoteTelegramContextBlock(msg.text, 1200)}`,
         const suppressed = telegramThinkSuppressedRequest(request);
         const requestTimeoutMs = Number.isFinite(suppressed.timeoutMs) && (suppressed.timeoutMs ?? 0) > 0 ? suppressed.timeoutMs : void 0;
         const jsonStartMs = Date.now();
+        const nativeOllamaRouter = diagnostics?.backendType === "ollama" && typeof backend.nativeOllamaChatCompletion === "function";
         try {
-          jsonModeResult = await this.telegramObservableInference(
+          jsonModeResult = nativeOllamaRouter ? await backend.nativeOllamaChatCompletion({
+            ...suppressed,
+            responseFormat: TELEGRAM_INTERACTION_DECISION_RESPONSE_FORMAT,
+            disableEmptyContentRecovery: true
+          }) : await this.telegramObservableInference(
             backend,
             {
               ...suppressed,
@@ -647338,7 +647420,7 @@ ${this.quoteTelegramContextBlock(msg.text, 1200)}`,
         }
         const plainStartMs = Date.now();
         try {
-          const plainResult = await this.telegramObservableInference(
+          const plainResult = nativeOllamaRouter ? await backend.nativeOllamaChatCompletion(suppressed) : await this.telegramObservableInference(
             backend,
             suppressed,
             inferenceKind,
@@ -648028,11 +648110,11 @@ ${retryText}`,
       }
       telegramRouterAutoModelEnabled() {
         const raw = (process.env["OMNIUS_TG_ROUTER_AUTO_MODEL"] ?? "").trim().toLowerCase();
-        return raw !== "0" && raw !== "false" && raw !== "off";
+        return raw === "1" || raw === "true" || raw === "on";
       }
       telegramRouterCandidateModels() {
         const raw = (process.env["OMNIUS_TG_ROUTER_MODEL_CANDIDATES"] ?? "").trim();
-        const candidates = raw ? raw.split(/[,\s]+/).map((part) => part.trim()).filter(Boolean) : TELEGRAM_DEFAULT_ROUTER_MODEL_CANDIDATES;
+        const candidates = raw ? raw.split(/[,\s]+/).map((part) => part.trim()).filter(Boolean) : [];
         return Array.from(new Set(candidates));
       }
       telegramRouterAllowThinkHeavyAutoModels() {
@@ -648042,16 +648124,10 @@ ${retryText}`,
       telegramRouterModelLooksThinkHeavy(name10) {
         return /\b(?:qwen3|qwq|deepseek-r1|r1-|reasoning)\b/i.test(name10);
       }
-      orderTelegramRouterCandidates(candidates) {
-        if (this.telegramRouterAllowThinkHeavyAutoModels()) return candidates;
-        const stable = candidates.filter((candidate) => !this.telegramRouterModelLooksThinkHeavy(candidate));
-        const thinkHeavy = candidates.filter((candidate) => this.telegramRouterModelLooksThinkHeavy(candidate));
-        return [...stable, ...thinkHeavy];
-      }
       normalizeOllamaModelNameForMatch(name10) {
         return name10.trim().toLowerCase().replace(/:latest$/, "");
       }
-      async fetchOllamaInstalledModelNames(baseUrl) {
+      async fetchOllamaInstalledModels(baseUrl) {
         const url = `${baseUrl.replace(/\/+$/, "")}/api/tags`;
         const timeoutFn = AbortSignal.timeout;
         const res = await fetch(url, {
@@ -648059,7 +648135,43 @@ ${retryText}`,
         });
         if (!res.ok) throw new Error(`ollama /api/tags returned HTTP ${res.status}`);
         const data = await res.json();
-        return Array.isArray(data.models) ? data.models.map((model) => typeof model.name === "string" ? model.name : "").filter(Boolean) : [];
+        return Array.isArray(data.models) ? data.models.map((model) => ({
+          name: typeof model.name === "string" ? model.name : "",
+          sizeBytes: typeof model.size === "number" ? model.size : void 0,
+          parameterSize: typeof model.details?.parameter_size === "string" ? model.details.parameter_size : void 0
+        })).filter((model) => Boolean(model.name)) : [];
+      }
+      telegramModelParameterBillions(model) {
+        const haystack = `${model.name} ${model.parameterSize ?? ""}`.toLowerCase();
+        const billion = haystack.match(/(\d+(?:\.\d+)?)\s*(?:b|bn)\b/);
+        if (billion) return Number(billion[1]);
+        const million = haystack.match(/(\d+(?:\.\d+)?)\s*m\b/);
+        if (million) return Number(million[1]) / 1e3;
+        return null;
+      }
+      scoreTelegramInstalledRouterModel(model) {
+        const name10 = model.name.toLowerCase();
+        if (/(?:embed|embedding|nomic|bge|e5-|clip|rerank|moondream|llava|vision|vl\b|minicpm|whisper|tts|sdxl|diffusion)/i.test(name10)) {
+          return Number.NEGATIVE_INFINITY;
+        }
+        const paramsB = this.telegramModelParameterBillions(model);
+        if (paramsB !== null && paramsB < TELEGRAM_ROUTER_AUTO_MIN_PARAMETERS_B) {
+          return Number.NEGATIVE_INFINITY;
+        }
+        if (paramsB === null && (model.sizeBytes ?? 0) < 5e9) {
+          return Number.NEGATIVE_INFINITY;
+        }
+        let score = 0;
+        if (paramsB !== null) score += paramsB * 10;
+        else score += Math.min(80, (model.sizeBytes ?? 0) / 1e9);
+        if (/qwen|huihui|qwq/i.test(name10)) score += 80;
+        else if (/deepseek|nemotron|llama|mistral|mixtral|command-r|devstral/i.test(name10)) score += 50;
+        else if (/gemma/i.test(name10)) score += 20;
+        if (/:latest$/i.test(model.name)) score += 1;
+        if (this.telegramRouterModelLooksThinkHeavy(model.name) && !this.telegramRouterAllowThinkHeavyAutoModels()) {
+          score -= 15;
+        }
+        return score;
       }
       async resolveTelegramRouterBackend(config) {
         const explicit = (process.env["OMNIUS_TG_ROUTER_MODEL"] ?? "").trim();
@@ -648071,17 +648183,20 @@ ${retryText}`,
             detail: "OMNIUS_TG_ROUTER_MODEL"
           };
         }
-        if (config.backendType !== "ollama" || !this.telegramRouterAutoModelEnabled()) {
+        if (config.backendType !== "ollama") {
           return {
             backend: new OllamaAgenticBackend(config.backendUrl, config.model, config.apiKey),
             model: config.model,
             source: "main"
           };
         }
-        const candidates = this.orderTelegramRouterCandidates(this.telegramRouterCandidateModels());
+        const autoModelEnabled = this.telegramRouterAutoModelEnabled();
+        const candidateFilter = this.telegramRouterCandidateModels();
+        const candidates = new Set(candidateFilter.map((candidate) => this.normalizeOllamaModelNameForMatch(candidate)));
         const cacheKey = `${config.backendUrl}
 ${config.model}
-${candidates.join(",")}`;
+auto=${autoModelEnabled ? "1" : "0"}
+${candidateFilter.join(",")}`;
         const now = Date.now();
         if (this.telegramRouterModelCache && this.telegramRouterModelCache.cacheKey === cacheKey && now - this.telegramRouterModelCache.atMs < 6e4) {
           const cached = this.telegramRouterModelCache;
@@ -648092,30 +648207,65 @@ ${candidates.join(",")}`;
             detail: cached.detail
           };
         }
+        if (!autoModelEnabled) {
+          const detail2 = "Telegram router auto-model selection is disabled by default; using main model";
+          this.telegramRouterModelCache = {
+            cacheKey,
+            atMs: now,
+            model: config.model,
+            source: "main",
+            detail: detail2
+          };
+          return {
+            backend: new OllamaAgenticBackend(config.backendUrl, config.model, config.apiKey),
+            model: config.model,
+            source: "main",
+            detail: detail2
+          };
+        }
         try {
-          const installed = await this.fetchOllamaInstalledModelNames(config.backendUrl);
+          const installed = await this.fetchOllamaInstalledModels(config.backendUrl);
           const installedByNormalized = /* @__PURE__ */ new Map();
-          for (const name10 of installed) {
-            installedByNormalized.set(this.normalizeOllamaModelNameForMatch(name10), name10);
+          for (const model of installed) {
+            installedByNormalized.set(this.normalizeOllamaModelNameForMatch(model.name), model);
           }
-          for (const candidate of candidates) {
-            const selected = installedByNormalized.get(this.normalizeOllamaModelNameForMatch(candidate));
-            if (!selected) continue;
+          const installedMain = installedByNormalized.get(this.normalizeOllamaModelNameForMatch(config.model));
+          if (installedMain) {
             const resolved = {
               cacheKey,
               atMs: now,
-              model: selected,
-              source: "auto-small",
-              detail: "selected first installed Telegram router candidate from Ollama /api/tags; think-heavy models are tried last unless OMNIUS_TG_ROUTER_ALLOW_THINK_MODELS=1"
+              model: installedMain.name,
+              source: "main",
+              detail: "main Telegram router model is installed in Ollama /api/tags; using main model by policy"
             };
             this.telegramRouterModelCache = resolved;
             return {
-              backend: new OllamaAgenticBackend(config.backendUrl, selected, config.apiKey),
-              model: selected,
-              source: "auto-small",
+              backend: new OllamaAgenticBackend(config.backendUrl, installedMain.name, config.apiKey),
+              model: installedMain.name,
+              source: "main",
               detail: resolved.detail
             };
           }
+          if (autoModelEnabled) {
+            const pool3 = candidateFilter.length > 0 ? installed.filter((model) => candidates.has(this.normalizeOllamaModelNameForMatch(model.name))) : installed;
+            const selected = pool3.map((model) => ({ model, score: this.scoreTelegramInstalledRouterModel(model) })).filter((entry) => Number.isFinite(entry.score)).sort((a2, b) => b.score - a2.score)[0]?.model;
+            if (selected) {
+              const resolved = {
+                cacheKey,
+                atMs: now,
+                model: selected.name,
+                source: "auto-installed",
+                detail: `main Telegram router model ${JSON.stringify(config.model)} was not found in Ollama /api/tags; selected best installed capable router model dynamically (minimum ${TELEGRAM_ROUTER_AUTO_MIN_PARAMETERS_B}B, excludes embeddings/vision/tiny models)`
+              };
+              this.telegramRouterModelCache = resolved;
+              return {
+                backend: new OllamaAgenticBackend(config.backendUrl, selected.name, config.apiKey),
+                model: selected.name,
+                source: "auto-installed",
+                detail: resolved.detail
+              };
+            }
+          }
         } catch (err) {
           const detail2 = `router model auto-detect failed: ${err instanceof Error ? err.message : String(err)}`;
           this.telegramRouterModelCache = {
@@ -648132,7 +648282,7 @@ ${candidates.join(",")}`;
             detail: detail2
           };
         }
-        const detail = "no configured small router model was installed; using main model";
+        const detail = `no installed capable Telegram router model met the dynamic minimum (${TELEGRAM_ROUTER_AUTO_MIN_PARAMETERS_B}B); using main model`;
         this.telegramRouterModelCache = {
           cacheKey,
           atMs: now,

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.209",
+  "version": "1.0.211",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.209",
+      "version": "1.0.211",
       "bundleDependencies": [
         "image-to-ascii"
       ],
@@ -2310,9 +2310,9 @@
       }
     },
     "node_modules/bare-path": {
-      "version": "3.0.0",
-      "resolved": "https://registry.npmjs.org/bare-path/-/bare-path-3.0.0.tgz",
-      "integrity": "sha512-tyfW2cQcB5NN8Saijrhqn0Zh7AnFNsnczRcuWODH0eYAXBsJ5gVxAUuNr7tsHSC6IZ77cA0SitzT+s47kot8Mw==",
+      "version": "3.0.1",
+      "resolved": "https://registry.npmjs.org/bare-path/-/bare-path-3.0.1.tgz",
+      "integrity": "sha512-ghj2DSK/2e99a1anTVPCV4m4YIYtrbXhfM7V3D7XZLOTsybnYyaJloymGqssQc8l/or0UoDyRtNQkmkEF/ysgQ==",
       "license": "Apache-2.0",
       "optional": true,
       "dependencies": {
@@ -6886,9 +6886,9 @@
       }
     },
     "node_modules/undici": {
-      "version": "7.26.0",
-      "resolved": "https://registry.npmjs.org/undici/-/undici-7.26.0.tgz",
-      "integrity": "sha512-3O9Tf67pGhgOv9jM35AbhkXAKi13f3oy3aE4CSgr+TckGeY+/iu97ZXN+J7DpHPzLbVApFd1IFhcnBjREYXYcg==",
+      "version": "7.27.0",
+      "resolved": "https://registry.npmjs.org/undici/-/undici-7.27.0.tgz",
+      "integrity": "sha512-+t2Z/GwkZQDtu00813aP66ygViGtPHKhhoFZpQKpKrE+9jIgES+Zw+mFNaDWOVRKiuJjuqKHzD3B1sfGg8+ZOQ==",
       "license": "MIT",
       "engines": {
         "node": ">=20.18.1"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.209",
+  "version": "1.0.211",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",