.env.example

# Copy to `.env` (gitignored). Both frontend and backend read this file —
# Vite loads `VITE_*` keys into the React build, the Flask backend
# loads everything via python-dotenv at boot.

# ─── Backend: AI provider ──────────────────────────────────────────────────
API_KEY=your-api-key-here
API_URL=https://integrate.api.nvidia.com/v1

# Optional per-tier model overrides
# MODEL_DEFAULT=qwen/qwen3.5-122b-a10b
# MODEL_FAST=meta/llama-3.1-8b-instruct
# MODEL_QUALITY=deepseek-ai/deepseek-v4-pro
# MODEL_QWEN_122B=qwen/qwen3.5-122b-a10b
# MODEL_GLM_5_1=z-ai/glm-5.1
# MODEL_DEEPSEEK_V4_PRO=deepseek-ai/deepseek-v4-pro
# MODEL_KIMI_2_5=moonshotai/kimi-k2.5
# MODEL_NEMOTRON=nvidia/nemotron-3-super-120b-a12b
# MODEL_LLAMA=meta/llama-3.3-70b-instruct
# MAX_TOKENS=32768
# MODEL_VISION=nvidia/llama-3.1-nemotron-nano-vl-8b-v1

# ─── Backend: server ────────────────────────────────────────────────────────
HOST=127.0.0.1
PORT=5000
DEBUG=0
# Set to 1 only if you genuinely need to bind 0.0.0.0 (e.g. behind a proxy).
ALLOW_PUBLIC_BIND=0

# Comma-separated CORS allowlist. Use `*` only for prototyping.
CORS_ORIGINS=http://localhost:5173,http://127.0.0.1:5173,http://localhost:5000,http://127.0.0.1:5000

# Cap incoming request bodies (default 64 MB).
# MAX_CONTENT_LENGTH_BYTES=67108864

# ─── Backend: optional rate limit ──────────────────────────────────────────
# RATE_LIMIT=on
# RATE_LIMIT_DEFAULT=60/minute;10/second
# RATE_LIMIT_STORAGE=memory://

# ─── Backend: misc ─────────────────────────────────────────────────────────
# PREFLIGHT_CACHE_SECS=15
# PDF_MAX_PAGES=100

# ─── Frontend (Vite picks up only VITE_*) ──────────────────────────────────
# Override the dev/build-time backend URL. Empty = same origin (proxy).
VITE_BACKEND_URL=