.env.example

############ ENV VARIABLES YOU MUST SET YOURSELF ############

# Path to the jixia executable.
# Build jixia first (see README), then point this to the compiled binary.
# Example: JIXIA_PATH = "/home/you/jixia/.lake/build/bin/jixia"
JIXIA_PATH = "/path/to/jixia/.lake/build/bin/jixia"

# System root of your Lean 4 installation.
# Run `lake env` inside your project and copy the LEAN_SYSROOT line.
# Example: LEAN_SYSROOT = "/home/you/.elan/toolchains/leanprover--lean4---v4.18.0"
LEAN_SYSROOT = "/path/to/.elan/toolchains/leanprover--lean4---vX.Y.Z"

# PostgreSQL connection string.
# Create a database first: createdb my_database_name
# Example: CONNECTION_STRING = "dbname=physlibsearch user=myuser password=mypassword"
CONNECTION_STRING = "dbname=physlibsearch user=YOUR_USER password=YOUR_PASSWORD"

# Google Gemini API key — used for embeddings, informalization, and query expansion.
# Get yours at https://aistudio.google.com/app/apikey
GEMINI_API_KEY = "YOUR_GEMINI_API_KEY"

################### DEFAULT ENV VARIABLES ###################

# Model used for informalization (one-time indexing step; quality matters).
# Recommended: gemini-3.1-pro  |  Free tier alternative: gemma-4-31b-it
GEMINI_MODEL = "gemini-3.1-pro-preview"

# Model used for HyDE query expansion (real-time per query; speed matters).
# Recommended: gemini-3-flash-preview
GEMINI_FAST_MODEL = "gemini-3-flash-preview"

# Embedding model. gemini-embedding-2-preview produces 3072-dim cosine embeddings.
GEMINI_EMBEDDING_MODEL = "gemini-embedding-2-preview"

# Path to the folder where ChromaDB files are stored.
CHROMA_PATH = "chroma"

# Optional: use a custom LLM endpoint (e.g. OpenRouter) instead of Gemini directly.
# If set, GEMINI_FAST_MODEL must be the model name accepted by that endpoint.
# LLM_API_KEY = "sk-or-..."
# LLM_BASE_URL = "https://openrouter.ai/api/v1"

# Comma-separated list of allowed CORS origins for the API server.
# ALLOWED_ORIGINS = "https://yourdomain.com,http://localhost:3000"

# Log settings.
LOG_FILENAME = ""
LOG_FILEMODE = "w"
LOG_LEVEL = "WARNING"

# Dry run — runs indexing without calling any external API or computing embeddings.
# Useful for testing the pipeline end-to-end without spending API quota.
# Values: "true" | "false"
DRY_RUN = "false"