open-agents-ai 0.187.454 → 0.187.455

package/dist/index.js

@@ -581217,14 +581217,14 @@ function sanitizeChatContent(raw) {
   return cleaned.join("\n").trim();
 }
 async function directChatBackend(opts) {
-  const { model, messages: messages2, stream, res, sessionId, ollamaUrl } = opts;
+  const { model, messages: messages2, stream, res, sessionId, ollamaUrl, extraFields } = opts;
   const cfg = loadConfig();
   const isVllm = cfg.backendType === "vllm";
   const cleanModel = model.replace(/^[a-z]+\//, "");
   const headers = { "Content-Type": "application/json" };
   if (cfg.apiKey) headers["Authorization"] = `Bearer ${cfg.apiKey}`;
   if (isVllm) {
-    const reqBody = JSON.stringify({ model: cleanModel, messages: messages2, stream });
+    const reqBody = JSON.stringify({ model: cleanModel, messages: messages2, stream, ...extraFields || {} });
     if (stream) {
       res.writeHead(200, {
         "Content-Type": "text/event-stream",
@@ -581292,12 +581292,27 @@ async function directChatBackend(opts) {
       return content;
     }
   } else {
+    const ef = extraFields || {};
+    const ollamaOpts = {};
+    if (typeof ef["temperature"] === "number") ollamaOpts["temperature"] = ef["temperature"];
+    if (typeof ef["top_p"] === "number") ollamaOpts["top_p"] = ef["top_p"];
+    if (typeof ef["max_tokens"] === "number") ollamaOpts["num_predict"] = ef["max_tokens"];
+    if (typeof ef["seed"] === "number") ollamaOpts["seed"] = ef["seed"];
+    if (typeof ef["frequency_penalty"] === "number") ollamaOpts["frequency_penalty"] = ef["frequency_penalty"];
+    if (typeof ef["presence_penalty"] === "number") ollamaOpts["presence_penalty"] = ef["presence_penalty"];
+    if (Array.isArray(ef["stop"]) || typeof ef["stop"] === "string") ollamaOpts["stop"] = ef["stop"];
+    const hasTools = Array.isArray(ef["tools"]) && ef["tools"].length > 0;
     const reqBody = JSON.stringify({
       model: cleanModel,
       messages: messages2,
       stream,
-      think: false,
-      options: {}
+      // Don't force think:false when the caller is using tool calling —
+      // thinking models often need their reasoning chain to choose a tool.
+      ...hasTools ? {} : { think: false },
+      ...hasTools ? { tools: ef["tools"] } : {},
+      ...ef["tool_choice"] !== void 0 ? { tool_choice: ef["tool_choice"] } : {},
+      ...ef["response_format"] !== void 0 ? { format: ef["response_format"] } : {},
+      options: ollamaOpts
     });
     if (stream) {
       res.writeHead(200, {
@@ -581353,7 +581368,20 @@ async function directChatBackend(opts) {
       if (result.status >= 400) throw new Error(`Backend HTTP ${result.status}: ${result.body.slice(0, 200)}`);
       const j = JSON.parse(result.body);
       const content = j?.message?.content || "";
-      if (!content) throw new Error("Backend returned empty message content");
+      const rawToolCalls = j?.message?.tool_calls;
+      const toolCalls = Array.isArray(rawToolCalls) ? rawToolCalls.map((tc, idx) => ({
+        id: tc.id || `call_${Math.random().toString(36).slice(2, 10)}`,
+        type: "function",
+        function: {
+          name: tc?.function?.name ?? tc?.name ?? "",
+          arguments: typeof tc?.function?.arguments === "string" ? tc.function.arguments : JSON.stringify(tc?.function?.arguments ?? tc?.arguments ?? {})
+        },
+        index: idx
+      })) : [];
+      if (!content && toolCalls.length === 0) {
+        throw new Error("Backend returned empty message (no content, no tool_calls)");
+      }
+      const finishReason = toolCalls.length > 0 ? "tool_calls" : "stop";
       const promptTokens = j?.prompt_eval_count ?? 0;
       const completionTokens = j?.eval_count ?? 0;
       jsonResponse(res, 200, {
@@ -581363,8 +581391,12 @@ async function directChatBackend(opts) {
         model: cleanModel,
         choices: [{
           index: 0,
-          message: { role: "assistant", content },
-          finish_reason: "stop"
+          message: {
+            role: "assistant",
+            content,
+            ...toolCalls.length > 0 ? { tool_calls: toolCalls } : {}
+          },
+          finish_reason: finishReason
         }],
         usage: {
           prompt_tokens: promptTokens,
@@ -582008,10 +582040,13 @@ async function handleV1ChatCompletions(req2, res, ollamaUrl) {
     }
     return;
   }
+  const callerProvidedThink = "think" in routedBody;
+  const callerProvidedTools = Array.isArray(routedBody["tools"]) && routedBody["tools"].length > 0;
+  const finalThink = callerProvidedThink ? routedBody["think"] : callerProvidedTools ? void 0 : false;
   const ollamaPayload = JSON.stringify({
     ...routedBody,
     stream,
-    think: false
+    ...finalThink !== void 0 ? { think: finalThink } : {}
   });
   if (stream) {
     res.writeHead(200, {
@@ -582059,7 +582094,17 @@ async function handleV1ChatCompletions(req2, res, ollamaUrl) {
               const delta = {};
               if (ollamaChunk.message.role) delta.role = ollamaChunk.message.role;
               if (ollamaChunk.message.content) delta.content = ollamaChunk.message.content;
-              if (ollamaChunk.message.tool_calls) delta.tool_calls = ollamaChunk.message.tool_calls;
+              if (ollamaChunk.message.tool_calls) {
+                delta.tool_calls = ollamaChunk.message.tool_calls.map((tc, idx) => ({
+                  id: tc.id || `call_${randomBytes21(8).toString("hex")}`,
+                  type: "function",
+                  function: {
+                    name: tc?.function?.name ?? tc?.name ?? "",
+                    arguments: typeof tc?.function?.arguments === "string" ? tc.function.arguments : JSON.stringify(tc?.function?.arguments ?? tc?.arguments ?? {})
+                  },
+                  index: idx
+                }));
+              }
               const sseEvent = {
                 id: chatId,
                 object: "chat.completion.chunk",
@@ -582135,7 +582180,15 @@ async function handleV1ChatCompletions(req2, res, ollamaUrl) {
         content: ollamaResp.message?.content ?? ""
       };
       if (ollamaResp.message?.tool_calls && ollamaResp.message.tool_calls.length > 0) {
-        responseMessage.tool_calls = ollamaResp.message.tool_calls;
+        responseMessage.tool_calls = ollamaResp.message.tool_calls.map((tc, idx) => ({
+          id: tc.id || `call_${randomBytes21(8).toString("hex")}`,
+          type: "function",
+          function: {
+            name: tc?.function?.name ?? tc?.name ?? "",
+            arguments: typeof tc?.function?.arguments === "string" ? tc.function.arguments : JSON.stringify(tc?.function?.arguments ?? tc?.arguments ?? {})
+          },
+          index: idx
+        }));
         if (!ollamaResp.message.content) responseMessage.content = null;
       }
       const hasToolCalls = !!ollamaResp.message?.tool_calls?.length;
@@ -584601,19 +584654,42 @@ data: ${JSON.stringify(data)}
       const session = getSession(sessionId, model, cwdPath);
       addUserMessage(session, chatBody.message);
       compactSession(session);
-      const wantsTools = chatBody["tools"] !== false && chatBody["use_tools"] !== false;
       const streamMode = chatBody.stream !== false;
-      if (!wantsTools) {
+      const toolsField = chatBody["tools"];
+      const isOpenAIToolsArray = Array.isArray(toolsField) && toolsField.length > 0 && toolsField.every(
+        (t2) => t2 && typeof t2 === "object" && t2["type"] === "function" && typeof t2["function"]?.["name"] === "string"
+      );
+      const wantsAgent = !isOpenAIToolsArray && chatBody["tools"] !== false && chatBody["use_tools"] !== false;
+      if (Array.isArray(chatBody.messages)) {
+        const incomingSystem = chatBody.messages.find((m2) => m2?.role === "system" && typeof m2.content === "string");
+        if (incomingSystem && session.messages.length > 0 && session.messages[0].role === "system") {
+          session.messages[0] = { ...session.messages[0], content: incomingSystem.content };
+        }
+      }
+      if (!wantsAgent) {
         try {
           const ans = await directChatBackend({
             model,
-            // Filter out tool_call/tool_result messages — the upstream
-            // model only accepts standard system/user/assistant turns.
             messages: session.messages.filter((m2) => m2.role === "system" || m2.role === "user" || m2.role === "assistant").map((m2) => ({ role: m2.role, content: m2.content })),
             stream: streamMode,
             res,
             sessionId: session.id,
-            ollamaUrl
+            ollamaUrl,
+            // Pass-through of OpenAI fields — the upstream model needs
+            // them for native tool calling, deterministic outputs, etc.
+            extraFields: {
+              ...isOpenAIToolsArray ? { tools: toolsField } : {},
+              ...chatBody["tool_choice"] !== void 0 ? { tool_choice: chatBody["tool_choice"] } : {},
+              ...chatBody["temperature"] !== void 0 ? { temperature: chatBody["temperature"] } : {},
+              ...chatBody["top_p"] !== void 0 ? { top_p: chatBody["top_p"] } : {},
+              ...chatBody["max_tokens"] !== void 0 ? { max_tokens: chatBody["max_tokens"] } : {},
+              ...chatBody["response_format"] !== void 0 ? { response_format: chatBody["response_format"] } : {},
+              ...chatBody["seed"] !== void 0 ? { seed: chatBody["seed"] } : {},
+              ...chatBody["frequency_penalty"] !== void 0 ? { frequency_penalty: chatBody["frequency_penalty"] } : {},
+              ...chatBody["presence_penalty"] !== void 0 ? { presence_penalty: chatBody["presence_penalty"] } : {},
+              ...chatBody["stop"] !== void 0 ? { stop: chatBody["stop"] } : {},
+              ...chatBody["parallel_tool_calls"] !== void 0 ? { parallel_tool_calls: chatBody["parallel_tool_calls"] } : {}
+            }
           });
           if (ans !== null) {
             addAssistantMessage(session, ans);

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "open-agents-ai",
-  "version": "0.187.454",
+  "version": "0.187.455",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "open-agents-ai",
-      "version": "0.187.454",
+      "version": "0.187.455",
       "hasInstallScript": true,
       "license": "CC-BY-NC-4.0",
       "dependencies": {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "open-agents-ai",
-  "version": "0.187.454",
+  "version": "0.187.455",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",