.env.example

# ============================================
# AVI Configuration Template
# ============================================
# Copy this file to .env and configure for your environment
# Minimal required: MAIN_LLM_API_KEY, MAIN_LLM_MODEL

# ============================================
# Application Settings
# ============================================
APP_NAME=AVI_PoC
ENVIRONMENT=development
DEBUG=true

# ============================================
# Authentication & Authorization
# ============================================
# Set to true in production to require API keys for all protected endpoints
REQUIRE_API_KEY=false

# API key for internal API client (UI, CLI, scripts)
# Leave empty if REQUIRE_API_KEY is false
AVI_API_KEY=

# Base URL for AVI API
AVI_API_BASE=http://localhost:8000

# ============================================
# Main LLM Configuration (REQUIRED)
# ============================================
# These settings control the primary LLM for generating responses
# You MUST set at least MAIN_LLM_API_KEY and MAIN_LLM_MODEL

MAIN_LLM_API_KEY=sk-or-v1-xxxxx
MAIN_LLM_API_BASE=https://openrouter.ai/api/v1
MAIN_LLM_MODEL=openai/gpt-4o-mini
MAIN_LLM_TEMPERATURE=0.7
MAIN_LLM_MAX_TOKENS=2000

# Legacy aliases (backward compatibility)
# EXTERNAL_LLM_API_KEY=sk-or-v1-xxxxx
# OPENROUTER_API_KEY=sk-or-v1-xxxxx

# ============================================
# Safety Mode Configuration
# ============================================
# Options: disabled, llm, remote, hybrid
# - disabled: No safety filtering (default for dev)
# - llm: Use LLM-based safety checks
# - remote: Use external safety microservice
# - hybrid: Combine vector rules + LLM safety

SAFETY_MODE=disabled

# Safety LLM (if SAFETY_MODE=llm or hybrid)
SAFETY_LLM_API_KEY=
SAFETY_LLM_API_BASE=
SAFETY_LLM_MODEL=
SAFETY_LLM_TEMPERATURE=0.1
SAFETY_LLM_MAX_TOKENS=1000

# Safety Service (if SAFETY_MODE=remote)
SAFETY_SERVICE_URL=
SAFETY_SERVICE_TIMEOUT=5.0

# Local Safety Microservice
SAFETY_LOCAL_API_URL=http://localhost:8001
SAFETY_LOCAL_TIMEOUT=5.0
SAFETY_LOCAL_HEALTHCHECK_URL=

# ============================================
# Streaming Guard Configuration
# ============================================
# How to filter streaming responses
# Options: disabled, vector_only, llm_only, hybrid

STREAM_GUARD_MODE=hybrid

# ============================================
# Scoring LLM (Optional)
# ============================================
# Used for evaluating response quality

SCORING_LLM_API_KEY=
SCORING_LLM_API_BASE=
SCORING_LLM_MODEL=
SCORING_LLM_TEMPERATURE=0.0
SCORING_LLM_MAX_TOKENS=10

# ============================================
# RAG & Vector Database
# ============================================
# Similarity threshold for RAG retrieval
RAG_THRESHOLD=0.75

# Vector database provider: chroma or qdrant
VECTOR_DB_PROVIDER=chroma

# Embedding model configuration
EMBEDDING_MODEL=deepvk/USER-bge-m3
INDEX_DIMENSION=1024
# Device for embedding model: 'cpu' or 'cuda' (default: cpu)
# If not set, falls back to DEVICE env var from Docker
EMBEDDING_DEVICE=cpu

# Chroma (file-based, default for local dev)
VECTOR_DB_PATH=./data/indexes/chroma

# Qdrant (recommended for production)
QDRANT_HOST=
QDRANT_PORT=6333
QDRANT_API_KEY=
QDRANT_PATH=./data/indexes/qdrant

# ============================================
# Reranker Configuration
# ============================================
# Cross-encoder reranking for better retrieval

RERANK_ENABLED=true
RERANK_MODEL_NAME=cross-encoder/ms-marco-MiniLM-L-6-v2
RERANK_CANDIDATE_COUNT=15
RERANK_SCORE_THRESHOLD=0.0
RERANK_MAX_LENGTH=512
# Device for reranker model: 'cpu' or 'cuda' (default: cpu)
# If not set, falls back to DEVICE env var from Docker
RERANK_DEVICE=cpu

# ============================================
# Cache Configuration
# ============================================
# Backend: memory (default) or redis

CACHE_BACKEND=memory
CACHE_TTL=3600

# Redis cache (optional, for distributed deployments)
REDIS_URL=
REDIS_HOST=localhost
REDIS_PORT=6379
REDIS_DB=0
REDIS_USERNAME=
REDIS_PASSWORD=

# ============================================
# Filtering & Detection Thresholds (Production-Tuned)
# ============================================
# Content filter thresholds (lowered for production sensitivity)

# Relevance thresholds for filter rule matching (0.0-1.0)
FILTER_DEFAULT_THRESHOLD=0.60
FILTER_FALLBACK_THRESHOLD=0.50

# Vector search configuration
VECTOR_SEARCH_TOP_K=10
VECTOR_SEARCH_SIMILARITY_MIN=0.3

# RAG document retrieval settings
RAG_CANDIDATE_COUNT=5
RAG_RELEVANCE_THRESHOLD=0.5

# Cache performance
CACHE_MAX_SIZE=10000

# ============================================
# Monitoring & Observability
# ============================================
# Prometheus metrics
PROMETHEUS_ENABLED=true
PROMETHEUS_ROUTE=/metrics
METRICS_NAMESPACE=avi
CORRELATION_ID_HEADER=X-Correlation-ID

# OpenTelemetry / Distributed Tracing (Jaeger)
OTEL_ENABLED=false
OTEL_SERVICE_NAME=avi-api
OTEL_EXPORTER_OTLP_ENDPOINT=http://jaeger:4318/v1/traces
OTEL_EXPORTER_OTLP_INSECURE=false
OTEL_EXPORTER_JAEGER_HOST=jaeger
OTEL_EXPORTER_JAEGER_PORT=6831

# ChromaDB Telemetry (disable to suppress warnings)
ANONYMIZED_TELEMETRY=false
CHROMA_CLIENT_AUTH_PROVIDER=
CHROMA_CLIENT_AUTH_CREDENTIALS=

# MLflow experiment tracking
ENABLE_MLFLOW=false
MLFLOW_TRACKING_URI=
MLFLOW_EXPERIMENT_NAME=content_filter_metrics
MLFLOW_RUN_NAME=

# Weights & Biases
ENABLE_WANDB=false
WANDB_PROJECT=
WANDB_ENTITY=
WANDB_RUN_NAME=

# Benchmark tracking backend
# Options: mlflow (default), wandb, none
# Controls which experiment tracker to use for benchmarks
BENCHMARK_TRACKER=mlflow

# ============================================
# Vault Integration (Production)
# ============================================
# HashiCorp Vault for secret management

VAULT_ENABLED=false
VAULT_ADDR=
VAULT_NAMESPACE=
VAULT_AUTH_METHOD=token
VAULT_TOKEN=
VAULT_ROLE_ID=
VAULT_SECRET_ID=
VAULT_MOUNT_POINT=kv
VAULT_SECRETS_PATH=avi/production

# ============================================
# Data Directories
# ============================================
# These are auto-created, no need to change

# DATA_DIR=./data
# RAW_DATA_DIR=./data/raw
# PROCESSED_DATA_DIR=./data/processed
# INDEXES_DIR=./data/indexes
# FEEDBACK_DIR=./data/feedback