oai responses api, tool call retries

server/src/llm/multiplexer.ts

@@ -1,7 +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 { buildToolLogMessageData, runToolAwareChatCompletions, runToolAwareOpenAIChat } from "./chat-tools.js";
 import { buildAnthropicConversationMessage, getAnthropicSystemPrompt } from "./message-content.js";
 import type { MultiplexRequest, MultiplexResponse, Provider } from "./types.js";
 
@@ -48,8 +48,8 @@ export async function runMultiplex(req: MultiplexRequest): Promise<MultiplexResp
     let raw: unknown;
     let toolMessages: ReturnType<typeof buildToolLogMessageData>[] = [];
 
-    if (req.provider === "openai" || req.provider === "xai") {
-      const client = req.provider === "openai" ? openaiClient() : xaiClient();
+    if (req.provider === "openai") {
+      const client = openaiClient();
       const r = await runToolAwareOpenAIChat({
         client,
         model: req.model,
@@ -66,6 +66,24 @@ export async function runMultiplex(req: MultiplexRequest): Promise<MultiplexResp
       outText = r.text;
       usage = r.usage;
       toolMessages = r.toolEvents.map((event) => buildToolLogMessageData(call.chatId, event));
+    } else if (req.provider === "xai") {
+      const client = xaiClient();
+      const r = await runToolAwareChatCompletions({
+        client,
+        model: req.model,
+        messages: req.messages,
+        temperature: req.temperature,
+        maxTokens: req.maxTokens,
+        logContext: {
+          provider: req.provider,
+          model: req.model,
+          chatId,
+        },
+      });
+      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();