# Copy to .env.local (gitignored). The daemon loads .env.local at startup.
# All optional — sensible defaults bake in. Real exported env vars override the file.

# --- Text brain (narration + chat). Default: local Ollama. ---
# PUCK_BRAIN_URL=http://127.0.0.1:11434/v1
# PUCK_BRAIN_MODEL=hf.co/liahmartens/Holotron-12B-GGUF:Q8_0
# PUCK_BRAIN_TIMEOUT=25

# --- Vision (screenshot → events). ---
# Cloud-primary (Modal vLLM) is the default stance: local 12B tiling melts laptops
# and fails on weak machines. Non-localhost URL → whole-image; localhost → tiled.
PUCK_VISION_URL=https://<you>--puck-brain-serve.modal.run/v1
PUCK_VISION_MODEL=Hcompany/Holotron-12B
# Optional local fallback (e.g. llama-server on :8080). Empty = cloud only.
PUCK_VISION_FALLBACK_URL=
PUCK_VISION_FALLBACK_MODEL=
# A cold vLLM 12B start can take minutes; once warm it's ~5-15s.
PUCK_VISION_TIMEOUT=120

# Modal auth (for `modal deploy server/brain_modal.py`): run `modal token new` — it
# writes ~/.modal.toml. It is NOT read from any env var; keep tokens out of the repo.