feat: support OpenRouter/custom LLM providers via env vars

Add LLM_API_KEY, LLM_BASE_URL, LLM_MODEL env vars so any
OpenAI-compatible API (OpenRouter, etc.) can be used as the LLM
backend without code changes.

src/lib/llm/providers/index.ts

@@ -13,9 +13,9 @@ export interface Provider {
 }
 
 export function getActiveProvider(): Provider {
+  const hasOpenAI = !!process.env.OPENAI_API_KEY || !!process.env.LLM_API_KEY;
   const hasTavily = !!process.env.TAVILY_API_KEY;
   const hasPerplexity = !!process.env.PERPLEXITY_API_KEY;
-  const hasOpenAI = !!process.env.OPENAI_API_KEY;
 
   if (hasTavily && hasPerplexity) {
     console.log("[llm] Using provider: Tavily search + Perplexity synthesis");

src/lib/llm/providers/openai.ts

@@ -8,10 +8,16 @@ import type {
 import type { SearchResult } from "./tavily";
 
 let _client: OpenAI | null = null;
+const MODEL = process.env.LLM_MODEL || "gpt-4o-mini";
 
 function getClient(): OpenAI {
   if (!_client) {
-    _client = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });
+    const baseURL = process.env.LLM_BASE_URL || undefined;
+    const apiKey = process.env.OPENAI_API_KEY || process.env.LLM_API_KEY;
+    if (!apiKey) {
+      throw new Error("No API key configured. Set OPENAI_API_KEY or LLM_API_KEY.");
+    }
+    _client = new OpenAI({ apiKey, baseURL });
   }
   return _client;
 }
@@ -112,7 +118,7 @@ Provide a comprehensive comparison with scores, pros/cons, and a recommendation.
   for (let attempt = 1; attempt <= MAX_RETRIES; attempt++) {
     try {
       const response = await getClient().chat.completions.create({
-        model: "gpt-4o-mini",
+        model: MODEL,
         messages: [
           { role: "system", content: SYSTEM_PROMPT },
           { role: "user", content: userPrompt },
@@ -179,7 +185,7 @@ Use the web research data above to provide factual, data-driven insights. Refere
   for (let attempt = 1; attempt <= MAX_RETRIES; attempt++) {
     try {
       const response = await getClient().chat.completions.create({
-        model: "gpt-4o-mini",
+        model: MODEL,
         messages: [
           { role: "system", content: SYSTEM_PROMPT },
           {