feat: centraliser le moteur IA via le Référentiel

server/_core/llm.ts

@@ -1,4 +1,5 @@
 import { ENV } from "./env";
+import { invokeSelectedReference } from "./aiReferenceClient";
 
 export type Role = "system" | "user" | "assistant" | "tool" | "function";
 
@@ -266,7 +267,6 @@ const normalizeResponseFormat = ({
 };
 
 export async function invokeLLM(params: InvokeParams): Promise<InvokeResult> {
-  assertApiKey();
 
   const {
     messages,
@@ -280,7 +280,6 @@ export async function invokeLLM(params: InvokeParams): Promise<InvokeResult> {
   } = params;
 
   const payload: Record<string, unknown> = {
-    model: "gemini-2.5-flash",
     messages: messages.map(normalizeMessage),
   };
 
@@ -312,21 +311,5 @@ export async function invokeLLM(params: InvokeParams): Promise<InvokeResult> {
     payload.response_format = normalizedResponseFormat;
   }
 
-  const response = await fetch(resolveApiUrl(), {
-    method: "POST",
-    headers: {
-      "content-type": "application/json",
-      authorization: `Bearer ${ENV.forgeApiKey}`,
-    },
-    body: JSON.stringify(payload),
-  });
-
-  if (!response.ok) {
-    const errorText = await response.text();
-    throw new Error(
-      `LLM invoke failed: ${response.status} ${response.statusText} – ${errorText}`
-    );
-  }
-
-  return (await response.json()) as InvokeResult;
+  return invokeSelectedReference<InvokeResult>(payload);
 }