VITE_MAPBOX_TOKEN=
VITE_MAPBOX_STYLE_URL=mapbox://styles/ownpath/cmqe4wg8h005001s4bjx9461m

HF_TOKEN=
HF_MODEL=Qwen/Qwen3-Coder-30B-A3B-Instruct
HF_CHAT_COMPLETIONS_URL=https://router.huggingface.co/v1/chat/completions
HF_REASONING_EFFORT=medium
HF_TIMEOUT_MS=30000
RESEARCH_ENABLED=1
OFFICIAL_DATA_ENABLED=1
RESEARCH_DEPTH=standard
RESEARCH_MAX_QUERIES=6
RESEARCH_RESULTS_PER_QUERY=3
RESEARCH_MAX_SOURCES=24
RESEARCH_TOTAL_TIMEOUT_MS=45000
SEARCH_PROVIDER=disabled
MCP_WEB_SEARCH_ENABLED=0
MCP_WEB_SEARCH_TIMEOUT_MS=6000
MCP_WEB_SEARCH_COMMAND=
MCP_WEB_SEARCH_ARGS=
MCP_WEB_SEARCH_TOOL=web_search
SERPAPI_API_KEY=
BRAVE_SEARCH_API_KEY=
TAVILY_API_KEY=
GOOGLE_SEARCH_API_KEY=
GOOGLE_SEARCH_CX=
GOOGLE_SEARCH_ENGINE_ID=
GOOGLE_MAPS_API_KEY=
GOOGLE_PLACES_API_KEY=
GOOGLE_ROUTES_API_KEY=
REDDIT_CLIENT_ID=
REDDIT_CLIENT_SECRET=
CREA_DDF_CLIENT_ID=
CREA_DDF_CLIENT_SECRET=

# auto = Nemotron (NVIDIA) first, HF model as fallback so the agent always runs.
AGENT_MODEL_PROVIDER=auto
# Free key from https://build.nvidia.com (nvapi-...). Required to run Nemotron.
NVIDIA_API_KEY=
# Hosted NIM id. The HF repo "...-BF16" has no inference provider and cannot be called.
NVIDIA_MODEL=nvidia/nemotron-3-nano-omni-30b-a3b-reasoning
NVIDIA_BASE_URL=https://integrate.api.nvidia.com/v1
NVIDIA_TIMEOUT_MS=60000
# llama.cpp runtime (hackathon-allowed): run a small GGUF locally via `llama-server`,
# e.g. an OpenBMB MiniCPM build, then set LLAMACPP_ENABLED=1.
#   llama-server -hf openbmb/MiniCPM4-8B-GGUF --port 8080
LLAMACPP_ENABLED=0
LLAMACPP_BASE_URL=http://127.0.0.1:8080/v1
LLAMACPP_MODEL=local-gguf
LLAMACPP_API_KEY=
LLAMACPP_TIMEOUT_MS=60000
# Agentic workflow toggles:
# AGENT_DISCOVER=1 -> the model discovers the neighbourhoods + coordinates (nothing hardcoded;
#   seed list is only a fallback). Needs a capable brain; falls back to HF for discovery.
# AGENT_FANOUT=1 -> per-agent multi-model fan-out: each City Agent reasons via the agentic model.
AGENT_DISCOVER=1
AGENT_FANOUT=1
# Two-tier brains: the MAIN agentic model (Nemotron when NVIDIA_API_KEY is set, else HF) makes
# the decisions; this small assistant summarises each agent's evidence. The 0.5B is tiny, so it
# often falls back to the main model — use openbmb/MiniCPM4-8B-GGUF for genuine summaries.
AGENT_SUMMARIZER_PROVIDER=llamacpp
AGENT_PORT=8787