.env.example

# =============================================================================
# QueryWise — Environment Variables
# =============================================================================
# Copy this file to .env and fill in your values:
#   cp .env.example .env
# =============================================================================

# -- Application --
ENVIRONMENT=development
DEBUG=false

# -- App Database (pgvector) --
# Stores metadata, glossary, embeddings, query history
DATABASE_URL=postgresql+asyncpg://querywise:querywise_dev@localhost:5432/querywise

# -- Security --
# Fernet encryption key for stored connection strings
# Generate with: python -c "from cryptography.fernet import Fernet; print(Fernet.generate_key().decode())"
ENCRYPTION_KEY=dev-encryption-key-change-in-production

# Secrets backend for encrypting connection strings at rest.
# env (default, Fernet) | aws | gcp | azure | vault (cloud backends are seams,
# register an implementation via app.core.secrets.register_secrets_backend)
SECRETS_BACKEND=env

# Allowed CORS origins (JSON list)
CORS_ORIGINS=["http://localhost:5173"]

# -- Observability --
LOG_LEVEL=INFO
# console (human-readable, default) | json (structured, for log aggregation)
LOG_FORMAT=console
# Expose Prometheus metrics at GET /metrics
ENABLE_METRICS=true
SERVICE_NAME=querywise-backend
# OpenTelemetry tracing (off by default). When enabled with no OTLP endpoint,
# spans print to stdout; set the endpoint to export to Jaeger/Tempo/Collector.
OTEL_ENABLED=false
# OTEL_EXPORTER_OTLP_ENDPOINT=http://jaeger:4318/v1/traces

# -- Background Jobs --
# inprocess (asyncio, default — single process) | arq (Redis-backed, multi-replica)
# For arq: install the extra (pip install -e ".[jobs]") and run a worker:
#   JOB_BACKEND=arq arq app.jobs.worker.WorkerSettings
JOB_BACKEND=inprocess
REDIS_URL=redis://localhost:6379/0

# -- LLM Configuration --
# Uncomment ONE provider block below. Switching providers will clear and
# regenerate embeddings automatically (metadata is preserved).

# Vector dimension: 1536 for OpenAI/Anthropic, 768 for Ollama nomic-embed-text
EMBEDDING_DIMENSION=1536


# ---- Option 1: Anthropic  ----
# DEFAULT_LLM_PROVIDER=anthropic
# DEFAULT_LLM_MODEL=claude-sonnet-4-5-20250929
# ANTHROPIC_API_KEY=
# OPENAI_API_KEY=                        # Required for embeddings (Anthropic has no embedding API)
# EMBEDDING_MODEL=text-embedding-3-small

# ---- Option 2: OpenAI ----
DEFAULT_LLM_PROVIDER=openai
DEFAULT_LLM_MODEL=gpt-5.2
OPENAI_API_KEY=
EMBEDDING_MODEL=text-embedding-3-small

# ---- Option 3: Ollama (fully local — no API keys needed) ----
# DEFAULT_LLM_PROVIDER=ollama
# OLLAMA_MODEL=llama3.1:8b
# OLLAMA_EMBEDDING_MODEL=nomic-embed-text
# EMBEDDING_DIMENSION=768
# Native Ollama on macOS (GPU-accelerated via Metal — recommended):
# OLLAMA_BASE_URL=http://host.docker.internal:11434
# Docker Ollama (CPU-only, start with: docker compose --profile ollama-docker up):
# OLLAMA_BASE_URL=http://ollama:11434

# ---- Option 4: Azure OpenAI (run the pipeline inside a customer VPC) ----
# DEFAULT_LLM_PROVIDER=azure_openai
# DEFAULT_LLM_MODEL=<your-chat-deployment-name>
# AZURE_OPENAI_ENDPOINT=https://<resource>.openai.azure.com
# AZURE_OPENAI_API_KEY=
# AZURE_OPENAI_API_VERSION=2024-10-21
# AZURE_OPENAI_DEPLOYMENT=<your-embedding-deployment-name>
# EMBEDDING_MODEL=<your-embedding-deployment-name>

# -- Query Defaults --
DEFAULT_QUERY_TIMEOUT_SECONDS=30
DEFAULT_MAX_ROWS=1000
MAX_RETRY_ATTEMPTS=3

# -- Rate Limiting --
MAX_QUERIES_PER_MINUTE=30
RATE_LIMIT_ENABLED=true

# -- Context Builder --
MAX_CONTEXT_TABLES=8
MAX_SAMPLE_QUERIES=3

# -- Sample Database Auto-Setup --
# Set to false to skip auto-creating the IFRS 9 sample connection on startup
AUTO_SETUP_SAMPLE_DB=true
# SAMPLE_DB_CONNECTION_STRING=postgresql://sample:sample_dev@sample-db:5432/sampledb

# -- Frontend (used in docker-compose) --
VITE_API_URL=http://localhost:8000