#!/usr/bin/env bash
# LLM API — install on a clean Apple Silicon Mac.
# Usage: ./scripts/install.sh [--root ~/llm-api] [--no-llama] [--no-web] [--test]
set -euo pipefail

ROOT="${LLM_API_ROOT:-$HOME/llm-api}"
WITH_LLAMA=1
WITH_WEB=1
RUN_TEST=0
for a in "$@"; do
  case "$a" in
    --root) shift; ROOT="$1";;
    --root=*) ROOT="${a#--root=}";;
    --no-llama) WITH_LLAMA=0;;
    --no-web) WITH_WEB=0;;
    --test) RUN_TEST=1;;
  esac
done
HERE="$(cd "$(dirname "$0")/.." && pwd)"
say() { printf "\033[1;34m==>\033[0m %s\n" "$*"; }
warn() { printf "\033[1;33m!!\033[0m %s\n" "$*"; }

# 1. platform checks -------------------------------------------------------
[ "$(uname -s)" = "Darwin" ] || { echo "macOS required"; exit 1; }
[ "$(uname -m)" = "arm64" ] || { echo "Apple Silicon (arm64) required"; exit 1; }
CHIP=$(sysctl -n machdep.cpu.brand_string)
MEM_GB=$(( $(sysctl -n hw.memsize) / 1024 / 1024 / 1024 ))
FREE_GB=$(df -g "$HOME" | awk 'NR==2 {print $4}')
say "Detected $CHIP, ${MEM_GB} GB unified memory, ${FREE_GB} GB free on $HOME"
[ "$MEM_GB" -ge 16 ] || warn "Less than 16 GB of memory: only small models will run."
[ "$FREE_GB" -ge 120 ] || warn "Less than 120 GB free: downloads stop at the MIN_FREE_DISK_GB reserve."
export PATH="/opt/homebrew/bin:$HOME/.local/bin:$PATH"

# 2. toolchain ---------------------------------------------------------------
xcode-select -p >/dev/null 2>&1 || { say "Installing Xcode Command Line Tools (a dialog may appear)"; xcode-select --install || true; }
command -v brew >/dev/null || { say "Installing Homebrew"; /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"; eval "$(/opt/homebrew/bin/brew shellenv)"; }
command -v uv >/dev/null || { say "Installing uv"; brew install uv; }
if [ "$WITH_WEB" = 1 ]; then
  command -v node >/dev/null || { say "Installing Node"; brew install node; }
  command -v pnpm >/dev/null || { say "Installing pnpm"; brew install pnpm; }
fi
if [ "$WITH_LLAMA" = 1 ]; then
  command -v llama-server >/dev/null || { say "Installing llama.cpp (Metal)"; brew install llama.cpp; }
fi

# 3. directories ---------------------------------------------------------------
say "Creating $ROOT"
mkdir -p "$ROOT"/{models/{mlx,gguf,embeddings,rerankers,vision,manifests},data,logs}
chmod 700 "$ROOT/data"

# 4. python env -----------------------------------------------------------------
cd "$HERE/server"
if [ ! -x .venv/bin/python ]; then
  say "Creating Python 3.13 virtualenv"
  uv venv --python 3.13 .venv
fi
say "Installing Python dependencies (MLX, MLX-LM, mlx-vlm, FastAPI…)"
uv pip install --python .venv/bin/python -q -e ".[mlx,test]" "mlx-vlm>=0.7" asgi-lifespan
.venv/bin/python - <<'PY'
import mlx.core as mx, mlx_lm
print(f"    mlx {mx.__version__} · mlx-lm {mlx_lm.__version__} · Metal device: {mx.default_device()}")
PY

# 5. web ------------------------------------------------------------------------
if [ "$WITH_WEB" = 1 ]; then
  say "Installing and building the dashboard"
  (cd "$HERE/web" && pnpm install --frozen-lockfile --silent && NEXT_TELEMETRY_DISABLED=1 pnpm build | tail -3)
fi

# 6. env file (never overwrite) ----------------------------------------------------
if [ ! -f "$ROOT/.env" ]; then
  say "Writing $ROOT/.env from .env.example (edit ADMIN_EMAIL / ADMIN_PASSWORD / HF_TOKEN)"
  sed "s#/Users/USERNAME/llm-api#$ROOT#" "$HERE/.env.example" > "$ROOT/.env"
  chmod 600 "$ROOT/.env"
else
  say "$ROOT/.env already exists — left untouched"
fi

# 7. database + scan ------------------------------------------------------------------
say "Initializing database and scanning $ROOT/models"
(cd "$ROOT" && "$HERE/server/.venv/bin/llm-api" scan | tail -5)

# 8. optional local inference test -------------------------------------------------------
if [ "$RUN_TEST" = 1 ]; then
  say "Downloading a tiny test model and running one generation"
  "$HERE/server/.venv/bin/hf" download mlx-community/Qwen3-0.6B-4bit --local-dir "$ROOT/models/mlx/qwen/Qwen3-0.6B-4bit" >/dev/null
  "$HERE/server/.venv/bin/python" -m mlx_lm generate --model "$ROOT/models/mlx/qwen/Qwen3-0.6B-4bit" --prompt "Say hello in one word." --max-tokens 8 | tail -3
fi

cat <<EOF

Done. Next steps:
  1. Edit $ROOT/.env (ADMIN_EMAIL, ADMIN_PASSWORD, HF_TOKEN, PUBLIC_URL)
  2. Start the API:        cd $ROOT && $HERE/server/.venv/bin/llm-api serve
  3. Start the dashboard:  cd $HERE/web && pnpm start -p 8301 -H 127.0.0.1
  4. Open http://127.0.0.1:8300 — or put a TLS reverse proxy in front for https://$(grep -E '^DOMAIN=' "$ROOT/.env" | cut -d= -f2)
  Process management: see docs/deployment.md (PM2 / launchd examples).
EOF