[feature] adds web_search and fetch_url tool calls

server/src/llm/multiplexer.ts

@@ -1,6 +1,7 @@
 import { performance } from "node:perf_hooks";
 import { prisma } from "../db.js";
 import { anthropicClient, openaiClient, xaiClient } from "./providers.js";
+import { buildToolLogMessageData, runToolAwareOpenAIChat } from "./chat-tools.js";
 import type { MultiplexRequest, MultiplexResponse, Provider } from "./types.js";
 
 function asProviderEnum(p: Provider) {
@@ -44,25 +45,26 @@ export async function runMultiplex(req: MultiplexRequest): Promise<MultiplexResp
     let outText = "";
     let usage: MultiplexResponse["usage"] | undefined;
     let raw: unknown;
+    let toolMessages: ReturnType<typeof buildToolLogMessageData>[] = [];
 
     if (req.provider === "openai" || req.provider === "xai") {
       const client = req.provider === "openai" ? openaiClient() : xaiClient();
-      const r = await client.chat.completions.create({
+      const r = await runToolAwareOpenAIChat({
+        client,
         model: req.model,
-        // OpenAI SDK has very specific message union types; our normalized schema is compatible.
-        messages: req.messages.map((m) => ({ role: m.role, content: m.content, name: m.name })) as any,
+        messages: req.messages,
         temperature: req.temperature,
-        max_tokens: req.maxTokens,
+        maxTokens: req.maxTokens,
+        logContext: {
+          provider: req.provider,
+          model: req.model,
+          chatId,
+        },
       });
-      raw = r;
-      outText = r.choices?.[0]?.message?.content ?? "";
-      usage = r.usage
-        ? {
-            inputTokens: r.usage.prompt_tokens,
-            outputTokens: r.usage.completion_tokens,
-            totalTokens: r.usage.total_tokens,
-          }
-        : undefined;
+      raw = r.raw;
+      outText = r.text;
+      usage = r.usage;
+      toolMessages = r.toolEvents.map((event) => buildToolLogMessageData(call.chatId, event));
     } else if (req.provider === "anthropic") {
       const client = anthropicClient();
 
@@ -100,16 +102,27 @@ export async function runMultiplex(req: MultiplexRequest): Promise<MultiplexResp
 
     const latencyMs = Math.round(performance.now() - t0);
 
-    // Store assistant message + call record
-    await prisma.$transaction([
-      prisma.message.create({
+    // Store tool activity (if any), assistant message, and call record.
+    await prisma.$transaction(async (tx) => {
+      if (toolMessages.length) {
+        await tx.message.createMany({
+          data: toolMessages.map((message) => ({
+            chatId: message.chatId,
+            role: message.role as any,
+            content: message.content,
+            name: message.name,
+            metadata: message.metadata as any,
+          })),
+        });
+      }
+      await tx.message.create({
         data: {
           chatId: call.chatId,
           role: "assistant" as any,
           content: outText,
         },
-      }),
-      prisma.llmCall.update({
+      });
+      await tx.llmCall.update({
         where: { id: call.id },
         data: {
           response: raw as any,
@@ -118,8 +131,8 @@ export async function runMultiplex(req: MultiplexRequest): Promise<MultiplexResp
           outputTokens: usage?.outputTokens,
           totalTokens: usage?.totalTokens,
         },
-      }),
-    ]);
+      });
+    });
 
     return {
       provider: req.provider,