.env.example

# LLM Runtime: "ollama" (default, local) or "hf_api" (free tier fallback)
LLM_RUNTIME=ollama

# Ollama settings (local, no API key needed)
OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=llama3

# Hugging Face Inference API (free tier fallback, ~1000 req/day)
HF_API_TOKEN= Add the key here
HF_MODEL=mistralai/Mistral-7B-Instruct-v0.2

# Embeddings (runs fully local, no API)
EMBEDDING_MODEL=all-MiniLM-L6-v2
EMBEDDING_SIMILARITY_THRESHOLD=0.85

# Storage
SQLITE_PATH=./memory_audit.db

# Alerts
SLACK_WEBHOOK_URL=
ALERT_CRITICAL_KEYS=refund_days,price,plan,policy_version

# API
API_HOST=0.0.0.0
API_PORT=8000

# Memory backend: "sqlite" (default, zero-cost) or "qdrant" (needs Docker)
MEMORY_BACKEND=sqlite
QDRANT_URL=http://localhost:6333