chore: switch to DeepSeek API + sentence-transformers for free deployment

- Replace OpenAI LLM with DeepSeek (deepseek-chat) via OpenAI-compatible client
- Replace OpenAI embeddings with local sentence-transformers (all-MiniLM-L6-v2)
- Pre-download embedding model during Docker build
- Add .gitignore

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/config.py

@@ -4,10 +4,11 @@ from dotenv import load_dotenv
 load_dotenv()
 
 TELEGRAM_BOT_TOKEN = os.environ["TELEGRAM_BOT_TOKEN"]
-OPENAI_API_KEY = os.environ["OPENAI_API_KEY"]
-OPENAI_MODEL = os.getenv("OPENAI_MODEL", "gpt-4o")
-OPENAI_EMBEDDING_MODEL = os.getenv("OPENAI_EMBEDDING_MODEL", "text-embedding-3-small")
+DEEPSEEK_API_KEY = os.environ["DEEPSEEK_API_KEY"]
+DEEPSEEK_MODEL = os.getenv("DEEPSEEK_MODEL", "deepseek-chat")
+DEEPSEEK_BASE_URL = "https://api.deepseek.com"
 ADMIN_TELEGRAM_ID = int(os.environ["ADMIN_TELEGRAM_ID"])
 DB_PATH = os.getenv("DB_PATH", "/app/data/second_brain.db")
 CHROMA_PATH = os.getenv("CHROMA_PATH", "/app/data/chromadb")
 UPLOAD_PATH = os.getenv("UPLOAD_PATH", "/app/data/uploads")
+EMBEDDING_MODEL = os.getenv("EMBEDDING_MODEL", "all-MiniLM-L6-v2")

src/handlers/user.py

@@ -2,13 +2,13 @@ from pathlib import Path
 from telegram import Update
 from telegram.ext import ContextTypes
 from openai import AsyncOpenAI
-from config import OPENAI_API_KEY, OPENAI_MODEL, ADMIN_TELEGRAM_ID
+from config import DEEPSEEK_API_KEY, DEEPSEEK_BASE_URL, DEEPSEEK_MODEL, ADMIN_TELEGRAM_ID
 import database as db
 import rag
 import document as doc_processor
 from guard import is_academic_topic, REJECT_MESSAGE
 
-_client = AsyncOpenAI(api_key=OPENAI_API_KEY)
+_client = AsyncOpenAI(api_key=DEEPSEEK_API_KEY, base_url=DEEPSEEK_BASE_URL)
 
 ACADEMIC_SYSTEM_PROMPT = (
     "Anda adalah asisten akademik personal untuk dosen di perguruan tinggi Indonesia. "
@@ -128,7 +128,7 @@ async def ask(update: Update, context: ContextTypes.DEFAULT_TYPE):
     await update.message.chat.send_action("typing")
 
     # Guard: hanya izinkan topik akademik
-    if not await is_academic_topic(question, _client, OPENAI_MODEL):
+    if not await is_academic_topic(question, _client, DEEPSEEK_MODEL):
         await update.message.reply_text(REJECT_MESSAGE)
         return
 
@@ -154,7 +154,7 @@ async def ask(update: Update, context: ContextTypes.DEFAULT_TYPE):
 
     try:
         response = await _client.chat.completions.create(
-            model=OPENAI_MODEL,
+            model=DEEPSEEK_MODEL,
             messages=messages,
             max_tokens=1500,
         )

src/rag.py

@@ -1,10 +1,10 @@
 import asyncio
 import chromadb
-from openai import AsyncOpenAI
-from config import CHROMA_PATH, OPENAI_API_KEY, OPENAI_EMBEDDING_MODEL
+from sentence_transformers import SentenceTransformer
+from config import CHROMA_PATH, EMBEDDING_MODEL
 
-_client = AsyncOpenAI(api_key=OPENAI_API_KEY)
 _chroma = chromadb.PersistentClient(path=CHROMA_PATH)
+_embedding_model = SentenceTransformer(EMBEDDING_MODEL)
 
 
 def _get_collection(name: str):
@@ -12,8 +12,8 @@ def _get_collection(name: str):
 
 
 async def _embed(text: str) -> list:
-    resp = await _client.embeddings.create(model=OPENAI_EMBEDDING_MODEL, input=text)
-    return resp.data[0].embedding
+    embedding = await asyncio.to_thread(_embedding_model.encode, text)
+    return embedding.tolist()
 
 
 async def add_memory(user_id: int, memory_id: int, content: str, scope: str = "private"):