env.example

# StillMe - Learning AI system with RAG foundation Environment Configuration

# =============================================================================
# API Configuration
# =============================================================================
DEEPSEEK_API_KEY=sk-REPLACE_ME_WITH_YOUR_DEEPSEEK_KEY
OPENAI_API_KEY=sk-REPLACE_ME_WITH_YOUR_OPENAI_KEY
ANTHROPIC_API_KEY=sk-ant-REPLACE_ME_WITH_YOUR_ANTHROPIC_KEY
# External Data APIs (Phase 1 - Weather & News)
# GNews API key for news provider (free tier: 100 requests/day)
# Get your key at: https://gnews.io/api
GNEWS_API_KEY=REPLACE_ME_WITH_YOUR_GNEWS_KEY
# Claude model for evaluation (default: claude-3-haiku-20240307 for cost-effective evaluation)
# Options: claude-3-haiku-20240307, claude-3-sonnet-20240229, claude-3-opus-20240229
CLAUDE_MODEL=claude-3-haiku-20240307

# =============================================================================
# Database Configuration
# =============================================================================
DATABASE_URL=sqlite:///./data/stillme.db

# =============================================================================
# Server Configuration
# =============================================================================
# PORT is set by Railway automatically, but can be overridden for local development
PORT=8000
BACKEND_HOST=localhost
BACKEND_PORT=8000
FRONTEND_PORT=8501

# =============================================================================
# Learning Configuration
# =============================================================================
LEARNING_ENABLED=true
AUTO_APPROVAL_THRESHOLD=0.8
TRUST_SCORE_THRESHOLD=0.7

# =============================================================================
# Notification Configuration
# =============================================================================
EMAIL_SMTP_SERVER=smtp.gmail.com
EMAIL_SMTP_PORT=587
EMAIL_USERNAME=your_email@gmail.com
EMAIL_PASSWORD=REPLACE_ME_WITH_YOUR_APP_PASSWORD
TELEGRAM_BOT_TOKEN=REPLACE_ME_WITH_YOUR_BOT_TOKEN
TELEGRAM_CHAT_ID=REPLACE_ME_WITH_YOUR_CHAT_ID

# =============================================================================
# Notification Settings
# =============================================================================
NOTIFY_LEARNING_SESSIONS=true
NOTIFY_ERRORS=true
NOTIFY_SYSTEM_EVENTS=true

# =============================================================================
# Security Configuration
# =============================================================================
SECRET_KEY=REPLACE_ME_WITH_SECRET_KEY
ENCRYPTION_KEY=REPLACE_ME_WITH_ENCRYPTION_KEY

# API Authentication
# Set this to protect sensitive endpoints (database reset, scheduler control, etc.)
# If not set, authentication is disabled (development only - NOT for production!)
STILLME_API_KEY=REPLACE_ME_WITH_API_KEY
# API URL for testing (default: http://localhost:8000 for local, Railway URL for production)
# Set this to test against Railway backend: https://stillme-backend-production.up.railway.app
STILLME_API_URL=http://localhost:8000

# CORS Configuration
# Comma-separated list of allowed origins (e.g., "https://yourdomain.com,https://app.yourdomain.com")
# If not set, defaults to localhost only (development)
# Example: CORS_ALLOWED_ORIGINS=https://stillme.example.com,https://dashboard.stillme.example.com
CORS_ALLOWED_ORIGINS=

# =============================================================================
# Validator Chain Configuration (v0.6.4+)
# =============================================================================
# Enable validator chain to reduce hallucinations
ENABLE_VALIDATORS=false
ENABLE_TONE_ALIGN=true
# Evidence overlap threshold (lowered from 0.08 to 0.01 to prevent false positives
# when LLM translates/summarizes content, reducing vocabulary overlap)
VALIDATOR_EVIDENCE_THRESHOLD=0.01
VALIDATOR_CITATION_REQUIRED=true
STILLME_TONE_STYLE=neutral

# =============================================================================
# Self-Correction Configuration (Cost-Benefit Logic)
# =============================================================================
# Self-correction mode for RewriteLLM:
# - "off": No rewrites (disable self-correction)
# - "light": Conservative (max 1 rewrite for medium quality, max 2 for low quality)
# - "aggressive": More rewrites (max 2 for medium quality, max 2 for low quality)
# Default: "light" (balanced cost and quality)
SELF_CORRECTION_MODE=light

# =============================================================================
# Step-Level Validation & Consistency Checks (Phase 1 - SSR)
# =============================================================================
# Enable step-level validation for multi-step reasoning
# Default: true (enabled)
ENABLE_STEP_LEVEL_VALIDATION=true

# Enable self-consistency checks
# Default: true (enabled)
ENABLE_CONSISTENCY_CHECKS=true

# Minimum number of steps to trigger step-level validation
# Default: 2 (only validate if 2+ steps detected)
STEP_VALIDATION_MIN_STEPS=2

# Confidence threshold for flagging unreliable steps
# Steps with confidence below this will be flagged
# Default: 0.5
STEP_CONFIDENCE_THRESHOLD=0.5

# =============================================================================
# Persistent Cache Configuration (Railway/Production)
# =============================================================================
# Path to persistent volume for HuggingFace/SentenceTransformer model cache
# This prevents re-downloading models on every restart/redeploy
# On Railway: Set this to match railway.json volume mount path
# Example: PERSISTENT_CACHE_PATH=/app/hf_cache
# Alternative: Use HF_CACHE_PATH for custom HuggingFace cache location
PERSISTENT_CACHE_PATH=/app/hf_cache
HF_CACHE_PATH=
# Optional: HuggingFace auth token (private models / higher rate limits)
# Do NOT log this value in application logs
# HF_TOKEN=hf_XXXXXXXXXXXXXXXXXXXX

# =============================================================================
# Response Cache Configuration (v0.8+)
# =============================================================================
# Enable/disable caching for LLM responses, RAG retrievals, and HTTP responses
# Default: all enabled (true)
# Set to "false" to disable specific cache type
ENABLE_LLM_CACHE=true
ENABLE_RAG_CACHE=true
ENABLE_HTTP_CACHE=true

# Cache TTL (Time To Live) in seconds
# LLM Response Cache: 1 hour (3600s) - responses are cached for 1 hour
# RAG Retrieval Cache: 6 hours (21600s) - RAG results are cached for 6 hours
# HTTP Response Cache: 5 minutes (300s) - API responses are cached for 5 minutes
CACHE_TTL_LLM=3600
CACHE_TTL_RAG=21600
CACHE_TTL_HTTP=300

# Redis Configuration (Optional - for persistent cache)
# If Redis is available, cache will be persistent across restarts
# If not set, uses in-memory cache (lost on restart)
# REDIS_HOST=localhost
# REDIS_PORT=6379
# REDIS_PASSWORD=
# REDIS_URL=redis://localhost:6379

# =============================================================================
# ChromaDB Configuration
# =============================================================================
# ChromaDB database persistence path
# On Railway: Use /app/data/vector_db (default, uses persistent volume)
# Custom path: Set CHROMA_DB_PATH to override default
# Example: CHROMA_DB_PATH=/app/chroma_data
# CRITICAL: Use absolute path for Railway persistence (relative paths don't persist)
CHROMA_DB_PATH=

# Force reset ChromaDB database on startup (useful for fixing schema mismatch errors)
# Set to "true" to automatically delete and recreate vector_db directory on startup
# WARNING: This will delete all vector database data!
# Example: FORCE_DB_RESET_ON_STARTUP=true
FORCE_DB_RESET_ON_STARTUP=false

# =============================================================================
# Development Configuration
# =============================================================================
DEBUG=false
LOG_LEVEL=INFO
ENVIRONMENT=production