omnius 1.0.210 → 1.0.211

package/dist/index.js

@@ -561829,6 +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;
@@ -564689,6 +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,
@@ -567332,6 +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,
@@ -572272,6 +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");
@@ -647266,8 +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,
@@ -647327,7 +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,

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.210",
+  "version": "1.0.211",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.210",
+      "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": "1.0.210",
+  "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",