.env.example

# LLM provider for the judge and dataset generation: openrouter | ollama | openai | anthropic
RAGPROOF_LLM_PROVIDER=openrouter
RAGPROOF_LLM_API_KEY=

# Judge model. Prefer a small, cheap model and validate it with 'ragproof calibrate'.
# Judge metrics are skipped when this is unset; deterministic metrics still run.
RAGPROOF_JUDGE_MODEL=

# Model used by 'ragproof generate' for dataset synthesis.
RAGPROOF_GEN_MODEL=

# Optional base URL override, e.g. a proxy or a remote Ollama host.
RAGPROOF_LLM_BASE_URL=

# Judge pricing in USD per million tokens. Used for cost reporting and the
# RAGPROOF_MAX_COST_USD budget. Leave at 0 for free local models.
RAGPROOF_JUDGE_INPUT_COST_PER_MTOK=0
RAGPROOF_JUDGE_OUTPUT_COST_PER_MTOK=0

# Target pipeline for the HTTP adapter. Adapter configs reference these by name;
# the values are never logged or stored.
RAGPROOF_TARGET_BASE_URL=
RAGPROOF_TARGET_API_KEY=

# Limits
RAGPROOF_MAX_CONCURRENCY=4
# Hard budget per run in USD. The run stops gracefully when the budget is reached.
# Requires the pricing variables above to be set for paid providers.
RAGPROOF_MAX_COST_USD=5.00