env.example

# ============================================================================
# Weaver - Environment Configuration
# ============================================================================
#
# Quick Start:
#   cp env.example .env
#   # Edit the REQUIRED section below, then:
#   cd app/backend && docker compose up -d
#   make run-backend   # Terminal 1
#   make run-frontend  # Terminal 2
#
# This single .env file is used by BOTH backend and frontend.
# The backend reads it directly; the frontend reads it via next.config.ts.
#
# ============================================================================


# ============================================================================
# REQUIRED — You must set these to run Weaver
# ============================================================================

# --- LLM API Key (pick one) ---
# Option 1: OpenRouter (recommended — one key for many models)
#   Get your key at: https://openrouter.ai/keys
OPENROUTER_API_KEY=

# Option 2: OpenAI directly
#   Get your key at: https://platform.openai.com/api-keys
# OPENAI_API_KEY=

# Option 3: Anthropic Claude (official API)
#   Get your key at: https://platform.claude.com/settings/keys
#   Docs: https://docs.anthropic.com/en/api/getting-started
# ANTHROPIC_API_KEY=

# --- Database ---
# Default works with `docker compose up -d` out of the box.
DATABASE_URL=postgresql://research_rag:research_rag_dev@localhost:5432/research_rag

# --- Redis (task queue) ---
# Default works with `docker compose up -d` out of the box.
REDIS_URL=redis://localhost:6379


# ============================================================================
# RECOMMENDED — Needed for specific features
# ============================================================================

# Google Gemini API (for PDF OCR and audio transcription)
#   Get your key at: https://aistudio.google.com/apikey
#   Free tier available. Required for: scanned PDF processing, video transcription.
GOOGLE_API_KEY=

# Authentication — choose one approach:
#
# Option A: Skip auth entirely (easiest for local dev)
AUTH_BYPASS_ENABLED=true
#
# Option B: Use Supabase auth (production / multi-user)
# AUTH_BYPASS_ENABLED=false
# NEXT_PUBLIC_SUPABASE_URL=https://your-project.supabase.co
# NEXT_PUBLIC_SUPABASE_ANON_KEY=your-anon-key
# SUPABASE_JWT_SECRET=your-jwt-secret


# ============================================================================
# OPTIONAL — Sensible defaults, change only if needed
# ============================================================================

# --- LLM Model Selection ---
LLM_MODEL=google/gemini-2.5-flash
EMBEDDING_MODEL=openai/text-embedding-3-small

# --- Local Deployment (optional — for running without cloud APIs) ---
# Uncomment these to use Ollama or another local LLM server.
# See docs/LOCAL_DEPLOYMENT.md for setup instructions.
#
# LLM_PROVIDER_TYPE=local               # openrouter (default) | openai | anthropic | local
# LOCAL_LLM_BASE_URL=http://localhost:11434/v1
# LOCAL_LLM_MODEL=llama3.2:3b
# LOCAL_LLM_API_KEY=ollama
#
# EMBEDDING_PROVIDER_TYPE=local          # openrouter (default) | openai | local
# EMBEDDING_DIMENSION=768               # Must match your embedding model (768 for nomic-embed-text)
# LOCAL_EMBEDDING_BASE_URL=http://localhost:11434/v1
# LOCAL_EMBEDDING_MODEL=nomic-embed-text

# --- RAG Configuration ---
# traditional: chunk-based retrieval (default)
# long_context: full document context (NotebookLM style)
# auto: choose based on document size
RAG_MODE=traditional
RETRIEVAL_TOP_K=5
RETRIEVAL_MIN_SIMILARITY=0.0

# Long context mode settings
LONG_CONTEXT_SAFETY_RATIO=0.55
LONG_CONTEXT_MIN_TOKENS=10000

# Citation settings
ENABLE_CITATION_GROUNDING=true
CITATION_FORMAT=both

# Intent classification (disabled by default for faster TTFB)
INTENT_CLASSIFICATION_ENABLED=false
INTENT_CACHE_ENABLED=true

# --- OCR Configuration ---
# auto: unstructured first, fallback to Gemini for scanned PDFs
# unstructured: lightweight parser (no PyTorch)
# gemini: Google Gemini Vision OCR (best quality, requires GOOGLE_API_KEY)
# docling: heavy parser with PyTorch (optional install)
OCR_MODE=auto
GEMINI_OCR_CONCURRENCY=3

# --- Vector Store ---
# pgvector: PostgreSQL extension (default, no extra service needed)
# qdrant: dedicated vector database (better for large-scale)
VECTOR_STORE_PROVIDER=pgvector

# Qdrant settings (only if VECTOR_STORE_PROVIDER=qdrant)
# QDRANT_URL=http://localhost:6333
# QDRANT_API_KEY=
# QDRANT_COLLECTION_NAME=document_chunks

# --- CORS ---
CORS_ORIGINS=http://localhost:3000,http://localhost:3001

# --- General ---
ENVIRONMENT=development
LOG_LEVEL=INFO


# ============================================================================
# VIDEO PLATFORMS — Only if you import from these sources
# ============================================================================

# --- YouTube ---
YOUTUBE_RATE_LIMIT_ENABLED=true
YOUTUBE_MIN_DELAY=2.0
YOUTUBE_MAX_DELAY=60.0
# YOUTUBE_PROXY_URL=               # HTTP/SOCKS proxy for IP bans
# YOUTUBE_COOKIES_PATH=            # Path to cookies.txt for restricted videos

# Gemini audio transcription max duration (minutes)
GEMINI_AUDIO_MAX_DURATION_MINUTES=60

# --- Bilibili ---
# Authentication is OPTIONAL. Works without it via web scraping.
# Only needed for VIP/restricted videos. Use a test account!
BILIBILI_RATE_LIMIT_ENABLED=true
BILIBILI_MIN_DELAY=2.0
BILIBILI_MAX_DELAY=60.0
BILIBILI_ENABLE_YTDLP=true
# BILIBILI_SESSDATA=
# BILIBILI_BILI_JCT=
# BILIBILI_BUVID3=

# --- Douyin/TikTok ---
# Get your key at: https://tikhub.io
# TIKHUB_API_KEY=
# TIKHUB_BASE_URL=https://api.tikhub.io

# Video storage path
VIDEO_STORAGE_PATH=data/videos


# ============================================================================
# PRODUCTION / CLOUD — Not needed for local development
# ============================================================================

# --- Supabase (production database + auth + storage) ---
# DATABASE_CLIENT_TYPE=supabase
# SUPABASE_URL=https://your-project.supabase.co
# SUPABASE_SERVICE_ROLE_KEY=your-service-role-key
# STORAGE_BUCKET=documents

# --- Observability ---
# Langfuse (LLM tracing): https://langfuse.com
LANGFUSE_ENABLED=false
# LANGFUSE_PUBLIC_KEY=
# LANGFUSE_SECRET_KEY=
# LANGFUSE_HOST=https://cloud.langfuse.com

# Loki (centralized logging)
LOKI_ENABLED=false
# LOKI_URL=http://localhost:3100/loki/api/v1/push

# --- Encryption ---
# For encrypting sensitive data in settings storage.
# Generate: python -c "from cryptography.fernet import Fernet; print(Fernet.generate_key().decode())"
# ENCRYPTION_KEY=

# --- URL Extraction ---
URL_EXTRACTION_TIMEOUT=60
URL_CONTENT_MAX_LENGTH=50000
DISABLE_SSRF_CHECK=false