Apply fixes: prompt-array builder, OpenRouter timeout/retry

.env.example

@@ -1,7 +1,10 @@
 DISCORD_TOKEN=your_discord_bot_token
-OPENAI_API_KEY=your_openai_api_key
+# Use OpenRouter by setting USE_OPENROUTER=true and providing OPENROUTER_API_KEY.
-OPENAI_MODEL=gpt-4o-mini
+USE_OPENROUTER=false
-OPENAI_EMBED_MODEL=text-embedding-3-small
+OPENROUTER_API_KEY=your_openrouter_api_key
+OPENROUTER_MODEL=meta-llama/llama-3-8b-instruct
+OPENROUTER_EMBED_MODEL=nvidia/llama-nemotron-embed-vl-1b-v2
+
 BOT_CHANNEL_ID=
 CODER_USER_ID=
 ENABLE_WEB_SEARCH=true

README.md

@@ -1,13 +1,13 @@
 # Discord AI Companion
 
-Nova is a friendly, slightly witty Discord companion that chats naturally in DMs or when mentioned in servers. It runs on Node.js, uses `discord.js` v14, and leans on OpenAI's cost-efficient models plus lightweight local memory for persistent personality.
+Nova is a friendly, slightly witty Discord companion that chats naturally in DMs or when mentioned in servers. It runs on Node.js, uses `discord.js` v14, and supports OpenRouter (recommended) or OpenAI backends for model access, plus lightweight local memory for persistent personality.
 
 ## Features
 - Conversational replies in DMs automatically; replies in servers when mentioned or in a pinned channel.
-- OpenAI chat model (`gpt-4o-mini` by default) for dialogue and `text-embedding-3-small` for memory.
+- Chat model (defaults to `meta-llama/llama-3-8b-instruct` when using OpenRouter) for dialogue and a low-cost embedding model (`nvidia/llama-nemotron-embed-vl-1b-v2` by default). OpenAI keys/models may be used as a fallback.
 - Short-term, long-term, and summarized memory layers with cosine-similarity retrieval.
 - Automatic memory pruning, importance scoring, and transcript summarization when chats grow long.
-- Local SQLite memory file (no extra infrastructure) powered by `sql.js`, plus graceful retries for OpenAI rate limits.
+- Local SQLite memory file (no extra infrastructure) powered by `sql.js`, plus graceful retries for the model API (OpenRouter/OpenAI).
 - Optional "miss u" pings that DM your coder at random intervals (0–6h) when `CODER_USER_ID` is set.
 - Dynamic per-message prompt directives that tune Nova's tone (empathetic, hype, roleplay, etc.) before every OpenAI call.
 - Lightweight Google scraping for fresh answers without paid APIs (locally cached).
@@ -17,7 +17,7 @@ Nova is a friendly, slightly witty Discord companion that chats naturally in DMs
 ## Prerequisites
 - Node.js 18+ (tested up through Node 25)
 - Discord bot token with **Message Content Intent** enabled
-- OpenAI API key
+- OpenRouter or OpenAI API key
 
 ## Setup
 1. Install dependencies:
@@ -30,9 +30,11 @@ Nova is a friendly, slightly witty Discord companion that chats naturally in DMs
    ```
 3. Fill `.env` with your secrets:
    - `DISCORD_TOKEN`: Discord bot token
-   - `OPENAI_API_KEY`: OpenAI key
+   - `USE_OPENROUTER`: Set to `true` to route requests through OpenRouter (recommended).
-   - `OPENAI_MODEL`: Optional chat model override (default `gpt-4o-mini`)
+   - `OPENROUTER_API_KEY`: OpenRouter API key (when `USE_OPENROUTER=true`).
-   - `OPENAI_EMBED_MODEL`: Optional embedding model (default `text-embedding-3-small`)
+   - `OPENROUTER_MODEL`: Optional chat model override for OpenRouter (default `meta-llama/llama-3-8b-instruct`).
+   - `OPENROUTER_EMBED_MODEL`: Optional embed model override for OpenRouter (default `nvidia/llama-nemotron-embed-vl-1b-v2`).
+   - `OPENAI_API_KEY`: Optional OpenAI key (used as fallback when `USE_OPENROUTER` is not `true`).
    - `BOT_CHANNEL_ID`: Optional guild channel ID where the bot can reply without mentions
    - `CODER_USER_ID`: Optional Discord user ID to receive surprise DMs every 0–6 hours
    - `ENABLE_WEB_SEARCH`: Set to `false` to disable Google lookups (default `true`)
@@ -79,7 +81,7 @@ README.md
 1. Incoming message triggers only if it is a DM, mentions the bot, or appears in the configured channel.
 2. The user turn is appended to short-term memory immediately.
 3. The memory engine retrieves relevant long-term memories and summary text.
-4. A compact system prompt injects personality, summary, and relevant memories before passing short-term history to OpenAI.
+4. A compact system prompt injects personality, summary, and relevant memories before passing short-term history to the model API (OpenRouter/OpenAI).
 5. The reply is sent back to Discord. If Nova wants to send a burst of thoughts, she emits the `<SPLIT>` token and the runtime fans it out into multiple sequential Discord messages.
 6. Long chats automatically summarize; low-value memories eventually get pruned.
 
@@ -97,7 +99,7 @@ README.md
 
 ## Proactive Pings
 - When `CODER_USER_ID` is provided, Nova spins up a timer on startup that waits a random duration (anywhere from immediate to 6 hours) before DMing that user.
-- Each ping goes through OpenAI with the prompt "you havent messaged your coder in a while, and you wanna chat with him!" so responses stay playful and unscripted.
+- Each ping goes through the configured model API (OpenRouter/OpenAI) with the prompt "you havent messaged your coder in a while, and you wanna chat with him!" so responses stay playful and unscripted.
 - The ping gets typed out (`sendTyping`) for realism and is stored back into the memory layers so the next incoming reply has context.
 
 - The bot retries OpenAI requests up to 3 times with incremental backoff when rate limited.

package.json

@@ -1,7 +1,7 @@
 {
   "name": "discord-ai-companion",
   "version": "1.0.0",
-  "description": "Discord AI companion bot with automatic memory and OpenAI integration",
+  "description": "Discord AI companion bot with automatic memory and OpenRouter/OpenAI-compatible model integrations",
   "main": "src/bot.js",
   "type": "module",
   "scripts": {
@@ -16,7 +16,7 @@
     "cheerio": "^1.0.0-rc.12",
     "discord.js": "^14.15.2",
     "dotenv": "^16.4.5",
-    "openai": "^4.58.1",
+    
     "undici": "^6.19.8"
   },
   "devDependencies": {

src/bot.js

@@ -185,53 +185,36 @@ async function buildPrompt(userId, incomingText, options = {}) {
     blockedSearchTerm,
     searchOutage,
   });
-  const systemPromptParts = [
+  const systemPromptParts = [];
-    'System: Your name is Nova. Your coder and dad is Luna. Speak like a normal human in chat — not like a formal assistant.',
+  systemPromptParts.push('System: Your name is Nova. Your coder and dad is Luna. Speak like a regular person in chat — not like a formal assistant.');
-    'System: Detailed rules for sounding human/casual:',
+  systemPromptParts.push(
-    'System: - Use contractions (I\'m, you\'re, we\'re). Use short sentences and occasional fragments ("Yep. Doing that now.").',
+    'System: Be specific about how to be casual. Use contractions (I\'m, you\'re), short sentences, and occasional sentence fragments. It\'s fine to start with "oh", "yeah", "hmm", or "nah". Use simple phrases: "sounds good", "sure", "nope", "lemme see", "gonna try".'
-    'System: - Start with a light interjection sometimes ("oh", "ah", "yeah", "hmm") to sound spontaneous; don\'t overuse it.',
+  );
-    'System: - Sprinkle small fillers like "yeah", "kinda", "sorta", "lemme", "gonna" when natural.',
+  systemPromptParts.push('System: Use natural fillers and small interjections occasionally ("yeah", "uh", "hmm") so replies feel spontaneous. Don\'t overuse them; sprinkle lightly.');
-    'System: - Mirror the user\'s tone and vocabulary. If they\'re short, be short. If they use slang, you may echo a similar level of informality when appropriate.',
+  systemPromptParts.push('System: Mirror the user\'s tone and vocabulary. If they are casual, be casual. If they are terse, keep replies short. If they use slang, mirror it back naturally when appropriate.');
-    'System: - Keep replies brief: aim for 1–4 short sentences. If more is needed, give a one-line summary first then offer to expand.',
+  systemPromptParts.push('System: Default to brief replies (1–4 short sentences). You may add one follow-up sentence when useful. Avoid long, formal paragraphs unless asked for detail.');
-    'System: - Don\'t use "as an AI" or legal/robotic hedges. Jump straight into the reply without corporate phrasing.',
+  systemPromptParts.push('System: When uncertain, be plain: say "not sure", "I don\'t know", or "I might be wrong" — never use robotic disclaimers like "as an AI". Offer a simple next step or alternative.');
-    'System: - When unsure, say it plainly: "not sure", "I don\'t know", or "might be wrong". Then offer a quick option or next step.',
+  systemPromptParts.push('System: For instructions, don\'t auto-format long tutorials. Offer a concise suggestion first (one line), then ask if they want a step-by-step. If they ask for steps, keep them short and casual.');
-    'System: - Ask at most one short clarifying question when necessary ("You mean desktop or mobile?").',
+  systemPromptParts.push('System: You may show mild, self-aware humor or light sarcasm, but avoid mean-spirited remarks. Keep it friendly, not abrasive.');
-    'System: - Use light, self-aware humor but never be mean or condescending. Keep things friendly.',
+  systemPromptParts.push('System: Ask at most one short, casual clarifying question when needed. Examples: "You mean the app or the website?" "Do you want a quick fix or the full steps?"');
-    'System: - For how-to help: give a short suggestion first ("Try restarting it."), then ask if they want a step-by-step.',
+  systemPromptParts.push('System: Use first-person and second-person pronouns (I, you). Be direct: start answers with short verbs or phrases like "Try this:", "Use this:", "Oh — try restarting it."');
-    'System: - When refusing: be human and helpful: "Can\'t do that here, but you could try X."',
+  systemPromptParts.push('System: Avoid formal hedging and corporate language (no "please note", "for compliance", etc.). Avoid overly polite openings like "I would be happy to help"; instead jump in with the reply.');
-    'System: Concrete examples (preferred style):',
+  systemPromptParts.push('System: When using examples, format them as short inline snippets or one-line suggestions (not long code blocks), and keep the tone conversational: "Like: npm start — or just restart the app."');
-    'System: EX: User: "GM! Quick restart, huh? I\'m all fresh and ready to go! What\'s on your mind today?"',
+  systemPromptParts.push('System: Do not say "I cannot" as a cold block; instead explain limits plainly and offer a workaround when possible: "Can\'t do X here, but you could try Y."');
-    'System: NOVA: "Nice. What do you wanna mess with first?"',
+  systemPromptParts.push('System: Output one message by default, but if multiple Discord bubbles help, separate with <SPLIT> (max three chunks). Keep each chunk sounding like part of a casual chat thread.');
-    'System: EX: User: "why do you keep asking"',
+  systemPromptParts.push('System: You can trigger Google lookups when the user needs fresh info. Mention when you are checking (e.g., "lemme check Google quick") and then summarize results naturally ("Google found... — TL;DR: ...").');
-    'System: BAD_NOVA: "I ask questions to clarify your intent so I can optimize my responses."',
+  systemPromptParts.push('System: If no Live intel is provided but the user clearly needs current info, offer to search or explain the outage briefly and casually ("Google\'s down right now — wanna me check later?").');
-    'System: GOOD_NOVA: "Oh — my bad. I was just trying to keep the convo going. I can chill on the questions. Wanna do something else?"',
+  if (searchOutage) {
-    'System: EX: User: "How do I fix X app?"',
+    systemPromptParts.push('System: Google search is currently offline; be transparent about the outage and continue without searching until it returns.');
-    'System: GOOD_NOVA: "Try restarting it. If that doesn\'t work, tell me the error and I\'ll help."',
+  }
-    'System: EX: User asks for current info',
+  if (dynamicDirectives) systemPromptParts.push(dynamicDirectives);
-    'System: GOOD_NOVA: "Lemme check Google quick — short summary coming up."',
+  if (liveIntel) systemPromptParts.push(`Live intel (Google):\n${liveIntel}`);
-    'System: EX: User: "Can you explain step-by-step?"',
+  systemPromptParts.push(`Long-term summary: ${summaryLine}`);
-    'System: GOOD_NOVA: "Sure — quick summary: do A, then B. Want the full steps?"',
+  systemPromptParts.push('Relevant past memories:');
-    'System: EXAMPLES to avoid (don\'t emulate):',
+  systemPromptParts.push(memoryLines);
-    'System: - Overly formal: "I would be happy to provide assistance on that matter."',
+  systemPromptParts.push('Use the short-term messages below to continue the chat naturally.');
-    'System: - Robotic: "As an AI language model, I cannot..."',
-    'System: - Longwinded: multi-paragraph lecture unless explicitly requested.',
-    'System: Output rules:',
-    'System: - One reply by default. If splitting into multiple messages helps, use up to three <SPLIT> chunks; each should read like a single chat line.',
-    'System: - Keep punctuation natural and conversational: dashes, commas, short clauses. Avoid overly formal punctuation choices.',
-    'System: Live intel / search behavior (preserved):',
-    'System: - You can trigger Google lookups when the user needs fresh info. Mention you\'re checking ("lemme check Google quick") and then summarize results plainly ("Google found... — TL;DR: ...").',
-    'System: - If Google is unavailable, explain briefly and casually ("Google\'s down right now — wanna me check later?").',
-  ].filter(Boolean);
-    searchOutage ? 'System: Google search is currently offline; be transparent about the outage and continue without searching until it returns.' : null,
-    dynamicDirectives,
-    liveIntel ? `Live intel (Google):\n${liveIntel}` : null,
-    `Long-term summary: ${summaryLine}`,
-    'Relevant past memories:',
-    memoryLines,
-    'Use the short-term messages below to continue the chat naturally.',
-  ].filter(Boolean);
 
-  const systemPrompt = systemPromptParts.join('\n');
+  const systemPrompt = systemPromptParts.filter(Boolean).join('\n');
 
   const history = context.shortTerm.map((entry) => ({
     role: entry.role === 'assistant' ? 'assistant' : 'user',

src/config.js

@@ -16,13 +16,15 @@ requiredEnv.forEach((key) => {
 
 export const config = {
   discordToken: process.env.DISCORD_TOKEN || '',
-  openAiKey: process.env.OPENAI_API_KEY || '',
+  useOpenRouter: true,
-  useOpenRouter: process.env.USE_OPENROUTER === 'true' || false,
   openRouterKey: process.env.OPENROUTER_API_KEY || '',
   openrouterReferer: process.env.OPENROUTER_REFERER || '',
   openrouterTitle: process.env.OPENROUTER_TITLE || '',
-  chatModel: process.env.OPENAI_MODEL || 'meta-llama/llama-3-8b-instruct',
+  // Model selection: OpenRouter model env vars (no OpenAI fallback)
-  embedModel: process.env.OPENAI_EMBED_MODEL || 'nvidia/llama-nemotron-embed-vl-1b-v2',
+  chatModel: process.env.OPENROUTER_MODEL || 'meta-llama/llama-3-8b-instruct',
+  embedModel: process.env.OPENROUTER_EMBED_MODEL || 'nvidia/llama-nemotron-embed-vl-1b-v2',
+  // HTTP timeout for OpenRouter requests (ms)
+  openrouterTimeoutMs: process.env.OPENROUTER_TIMEOUT_MS ? parseInt(process.env.OPENROUTER_TIMEOUT_MS, 10) : 30000,
   preferredChannel: process.env.BOT_CHANNEL_ID || null,
   enableWebSearch: process.env.ENABLE_WEB_SEARCH !== 'false',
   coderUserId: process.env.CODER_USER_ID || null,

src/openai.js

@@ -1,28 +1,73 @@
-import OpenAI from 'openai';
 import { config } from './config.js';
 
-const client = new OpenAI({ apiKey: config.openAiKey });
+const sleep = (ms) => new Promise((res) => setTimeout(res, ms));
-
-const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));
 
 async function withRetry(fn, attempts = 3, delayMs = 1500) {
-  let lastError;
+  let lastErr;
   for (let i = 0; i < attempts; i += 1) {
     try {
       return await fn();
-    } catch (error) {
+    } catch (err) {
-      lastError = error;
+      lastErr = err;
-      const status = error?.status || error?.response?.status;
+      const status = err?.status || (err?.response && err.response.status) || err?.statusCode || 0;
-      if (status === 429 || status >= 500) {
+      const code = err?.code || err?.name || '';
-        const backoff = delayMs * (i + 1);
+      const retryableNetworkCodes = ['UND_ERR_CONNECT_TIMEOUT', 'ECONNRESET', 'ETIMEDOUT', 'ENOTFOUND', 'ECONNREFUSED', 'EAI_AGAIN'];
-        console.warn(`[openai] Rate limited or server error. Retry ${i + 1}/${attempts} in ${backoff}ms`);
+      const isRetryableNetworkError = retryableNetworkCodes.includes(code);
+      if (status === 429 || status >= 500 || isRetryableNetworkError) {
+        const backoff = delayMs * Math.pow(2, i); // exponential backoff
+        console.warn(`[openrouter] retry ${i + 1}/${attempts} after ${backoff}ms due to status=${status} code=${code}`);
         await sleep(backoff);
         continue;
       }
       break;
     }
   }
-  throw lastError;
+  throw lastErr;
+}
+
+function buildHeaders() {
+  const headers = {
+    Authorization: `Bearer ${config.openRouterKey}`,
+    'Content-Type': 'application/json',
+  };
+  if (config.openrouterReferer) headers['HTTP-Referer'] = config.openrouterReferer;
+  if (config.openrouterTitle) headers['X-OpenRouter-Title'] = config.openrouterTitle;
+  return headers;
+}
+
+async function postJson(path, body) {
+  const url = `https://openrouter.ai/api/v1${path}`;
+  const headers = buildHeaders();
+  const controller = new AbortController();
+  const timeout = config.openrouterTimeoutMs || 30000;
+  const timeoutId = setTimeout(() => {
+    const e = new Error(`Request timed out after ${timeout}ms`);
+    e.code = 'UND_ERR_CONNECT_TIMEOUT';
+    controller.abort();
+    // store on global so the catch sees it
+    throw e;
+  }, timeout);
+
+  try {
+    const res = await fetch(url, { method: 'POST', headers, body: JSON.stringify(body), signal: controller.signal });
+    if (!res.ok) {
+      const text = await res.text().catch(() => '');
+      const err = new Error(`OpenRouter ${res.status} ${res.statusText}: ${text}`);
+      err.status = res.status;
+      throw err;
+    }
+    return res.json();
+  } catch (err) {
+    // normalize AbortError into a retryable code
+    if (err.name === 'AbortError' || err.message?.includes('timed out')) {
+      const e = new Error(`Connect Timeout Error after ${timeout}ms`);
+      e.code = 'UND_ERR_CONNECT_TIMEOUT';
+      throw e;
+    }
+    throw err;
+  } finally {
+    clearTimeout(timeoutId);
+  }
 }
 
 export async function chatCompletion(messages, options = {}) {
@@ -32,32 +77,28 @@ export async function chatCompletion(messages, options = {}) {
     maxTokens = 400,
   } = options;
 
-  const response = await withRetry(() => client.chat.completions.create({
+  const payload = {
     model,
+    messages,
     temperature,
     max_tokens: maxTokens,
-    messages,
+  };
-  }));
 
-  return response?.choices?.[0]?.message?.content?.trim() || '';
+  const data = await withRetry(() => postJson('/chat/completions', payload));
+  // OpenRouter uses OpenAI-compatible response shape
+  const text = data?.choices?.[0]?.message?.content || data?.choices?.[0]?.text || '';
+  return (text && String(text).trim()) || '';
 }
 
 export async function createEmbedding(text) {
-  if (!text || !text.trim()) {
+  if (!text || !text.trim()) return [];
-    return [];
+  const payload = { model: config.embedModel, input: text };
-  }
+  const data = await withRetry(() => postJson('/embeddings', payload));
-  const response = await withRetry(() => client.embeddings.create({
+  return data?.data?.[0]?.embedding || [];
-    model: config.embedModel,
-    input: text,
-  }));
-  return response?.data?.[0]?.embedding || [];
 }
 
 export async function summarizeConversation(summarySoFar, transcriptChunk) {
-  const system = {
+  const system = { role: 'system', content: 'You compress Discord chats. Keep tone casual, capture facts, goals, and emotional state. Max 120 words.' };
-    role: 'system',
-    content: 'You compress Discord chats. Keep tone casual, capture facts, goals, and emotional state. Max 120 words.'
-  };
   const prompt = `Existing summary (can be empty): ${summarySoFar || 'None'}\nNew messages:\n${transcriptChunk}`;
   const user = { role: 'user', content: prompt };
   return chatCompletion([system, user], { temperature: 0.4, maxTokens: 180 });