Feat: add env-driven LLM configuration and smoke test

server/_core/llm.ts

@@ -210,13 +210,13 @@ const normalizeToolChoice = (
 };
 
 const resolveApiUrl = () =>
-  ENV.forgeApiUrl && ENV.forgeApiUrl.trim().length > 0
-    ? `${ENV.forgeApiUrl.replace(/\/$/, "")}/v1/chat/completions`
+  ENV.llmApiUrl && ENV.llmApiUrl.trim().length > 0
+    ? ENV.llmApiUrl
     : "https://forge.manus.im/v1/chat/completions";
 
 const assertApiKey = () => {
-  if (!ENV.forgeApiKey) {
-    throw new Error("OPENAI_API_KEY is not configured");
+  if (!ENV.llmApiKey) {
+    throw new Error("LLM_API_KEY is not configured");
   }
 };
 
@@ -280,7 +280,7 @@ export async function invokeLLM(params: InvokeParams): Promise<InvokeResult> {
   } = params;
 
   const payload: Record<string, unknown> = {
-    model: "gemini-2.5-flash",
+    model: ENV.llmModel,
     messages: messages.map(normalizeMessage),
   };
 
@@ -296,9 +296,12 @@ export async function invokeLLM(params: InvokeParams): Promise<InvokeResult> {
     payload.tool_choice = normalizedToolChoice;
   }
 
-  payload.max_tokens = 32768
-  payload.thinking = {
-    "budget_tokens": 128
+  payload.max_tokens = ENV.llmMaxTokens;
+
+  if (ENV.llmEnableThinking && ENV.llmThinkingBudget > 0) {
+    payload.thinking = {
+      budget_tokens: ENV.llmThinkingBudget,
+    };
   }
 
   const normalizedResponseFormat = normalizeResponseFormat({
@@ -316,7 +319,7 @@ export async function invokeLLM(params: InvokeParams): Promise<InvokeResult> {
     method: "POST",
     headers: {
       "content-type": "application/json",
-      authorization: `Bearer ${ENV.forgeApiKey}`,
+      authorization: `Bearer ${ENV.llmApiKey}`,
     },
     body: JSON.stringify(payload),
   });