# --- Required secrets (fill these in) ---
GROQ_API_KEY=
TAVILY_API_KEY=
# Required when LLM_PROVIDER=gemini (default). Free key: aistudio.google.com
GOOGLE_API_KEY=
# HF token (read) for fetching task files from the gated GAIA dataset, since the
# scoring API's /files endpoint 404s. Accept terms at hf.co/datasets/gaia-benchmark/GAIA
HF_TOKEN=

# --- LLM provider for text/reasoning + judge nodes ---
# "openai_compatible" (freellmapi gateway, best for rate limits), "gemini", "groq".
LLM_PROVIDER=gemini
GEMINI_TEXT_MODEL=gemini-2.5-flash
# Image understanding: "gemini" (best), "groq", or "openai_compatible"
VISION_PROVIDER=gemini
GEMINI_VISION_MODEL=gemini-2.0-flash
GROQ_TEXT_MODEL=llama-3.1-8b-instant
# Global pace across all text-LLM calls (req/sec). 0.15 = ~9/min (<10 RPM free).
# Ignored when LLM_PROVIDER=openai_compatible (gateway manages its own limits).
RATE_LIMIT_RPS=0.15

# --- freellmapi gateway (set LLM_PROVIDER=openai_compatible to use) ---
OPENAI_COMPATIBLE_BASE_URL=
OPENAI_COMPATIBLE_API_KEY=
OPENAI_COMPATIBLE_MODEL=auto

# --- Optional: LangSmith tracing (nice for debugging the graph) ---
# Set TRACING=true and provide the API key to stream runs to smith.langchain.com.
# LangChain/LangGraph pick these up automatically; no code change needed.
LANGSMITH_TRACING=false
LANGSMITH_API_KEY=
LANGSMITH_PROJECT=gaia-agent
# LANGSMITH_ENDPOINT=https://api.smith.langchain.com  # set to EU host if needed

# --- Optional overrides (sane defaults applied if omitted) ---
# Groq text/reasoning model used by the agent + judge nodes
GROQ_TEXT_MODEL=llama-3.3-70b-versatile
# Groq multimodal model used by the describe_image tool
GROQ_VISION_MODEL=meta-llama/llama-4-scout-17b-16e-instruct
# Groq speech-to-text model used by the transcribe_audio tool
GROQ_WHISPER_MODEL=whisper-large-v3

# GAIA scoring API base URL
GAIA_API_URL=https://agents-course-unit4-scoring.hf.space

# How many times the judge may bounce a wrong answer back to research
MAX_JUDGE_RETRIES=2
# Hard cap on the LangGraph step budget per question
RECURSION_LIMIT=40
# Per-question wall-clock timeout (seconds); agent returns best-effort on expiry
QUESTION_TIMEOUT=180