secubox-openwrt/package/secubox/secubox-app-localai/files/usr/sbin/localaictl

#!/bin/sh
# SecuBox LocalAI Controller
# Copyright (C) 2025 CyberMind.fr
#
# LocalAI native binary management

CONFIG="localai"
BINARY="/usr/bin/local-ai"
DATA_DIR="/srv/localai"
BACKEND_ASSETS="/usr/share/localai/backend-assets"
LOCALAI_VERSION="3.9.0"

usage() {
	cat <<'EOF'
Usage: localaictl <command>

Install Commands:
  install            Download LocalAI binary from GitHub
  uninstall          Remove LocalAI binary

Service Commands:
  start              Start LocalAI service
  stop               Stop LocalAI service
  restart            Restart LocalAI service
  status             Show service status
  logs               Show logs (use -f to follow)

Model Commands:
  models             List installed models
  model-install <n>  Install model from preset or URL
  model-remove <n>   Remove installed model

Backend Commands:
  backends           List available backends

API Endpoints (default port 8081):
  /v1/models         - List models
  /v1/chat/completions - Chat completion
  /v1/completions    - Text completion
  /readyz            - Health check

Configuration: /etc/config/localai
EOF
}

require_root() { [ "$(id -u)" -eq 0 ] || { echo "Root required" >&2; exit 1; }; }

log_info() { echo "[INFO] $*"; logger -t localai "$*"; }
log_warn() { echo "[WARN] $*" >&2; logger -t localai -p warning "$*"; }
log_error() { echo "[ERROR] $*" >&2; logger -t localai -p err "$*"; }

uci_get() { uci -q get ${CONFIG}.$1; }
uci_set() { uci set ${CONFIG}.$1="$2" && uci commit ${CONFIG}; }

load_config() {
	api_port="$(uci_get main.api_port || echo 8081)"
	api_host="$(uci_get main.api_host || echo 0.0.0.0)"
	data_path="$(uci_get main.data_path || echo $DATA_DIR)"
	models_path="$(uci_get main.models_path || echo $DATA_DIR/models)"
	threads="$(uci_get main.threads || echo 4)"
	context_size="$(uci_get main.context_size || echo 2048)"

	mkdir -p "$data_path" "$models_path"
}

# =============================================================================
# INSTALL/UNINSTALL
# =============================================================================

get_arch() {
	local arch=$(uname -m)
	case "$arch" in
		aarch64) echo "arm64" ;;
		x86_64) echo "amd64" ;;
		*) echo "" ;;
	esac
}

cmd_install() {
	require_root
	load_config

	local arch=$(get_arch)
	if [ -z "$arch" ]; then
		log_error "Unsupported architecture: $(uname -m)"
		return 1
	fi

	if [ -x "$BINARY" ]; then
		log_warn "LocalAI already installed at $BINARY"
		local ver=$("$BINARY" run --help 2>&1 | grep -i version | head -1 || echo "installed")
		log_info "Status: $ver"
		echo ""
		echo "To reinstall, run: localaictl uninstall && localaictl install"
		return 0
	fi

	# LocalAI v2.x binary URL format
	local url="https://github.com/mudler/LocalAI/releases/download/v${LOCALAI_VERSION}/local-ai-Linux-${arch}"

	log_info "Downloading LocalAI v${LOCALAI_VERSION} for ${arch}..."
	log_info "URL: $url"
	echo ""

	# Create temp file
	local tmp_file="/tmp/local-ai-download"
	rm -f "$tmp_file"

	# Use -L to follow redirects (GitHub uses redirects)
	if wget -L --show-progress -O "$tmp_file" "$url" 2>&1; then
		# Verify it's an ELF binary by checking magic bytes
		local magic=$(head -c 4 "$tmp_file" | hexdump -e '4/1 "%02x"' 2>/dev/null)
		if [ "$magic" = "7f454c46" ]; then
			mv "$tmp_file" "$BINARY"
			chmod +x "$BINARY"
			log_info "LocalAI installed: $BINARY"

			# Mark as installed in config
			uci_set main.installed 1

			echo ""
			log_info "Binary downloaded successfully!"
			echo ""
			echo "To start the service:"
			echo "  uci set localai.main.enabled=1"
			echo "  uci commit localai"
			echo "  /etc/init.d/localai enable"
			echo "  /etc/init.d/localai start"
			echo ""
			echo "To download a model:"
			echo "  localaictl model-install tinyllama"
		else
			log_error "Downloaded file is not a valid ELF binary"
			rm -f "$tmp_file"
			return 1
		fi
	else
		log_error "Failed to download LocalAI"
		rm -f "$tmp_file"
		return 1
	fi
}

cmd_uninstall() {
	require_root

	if [ -x "$BINARY" ]; then
		# Stop service first
		/etc/init.d/localai stop 2>/dev/null

		rm -f "$BINARY"
		uci_set main.installed 0
		uci_set main.enabled 0
		log_info "LocalAI binary removed"
	else
		log_warn "LocalAI not installed"
	fi
}

# =============================================================================
# SERVICE MANAGEMENT
# =============================================================================

is_running() {
	pgrep -f "$BINARY" >/dev/null 2>&1 || pgrep -f "local-ai" >/dev/null 2>&1
}

cmd_start() {
	require_root
	load_config

	if ! [ -x "$BINARY" ]; then
		log_error "LocalAI binary not found: $BINARY"
		log_error "Run: localaictl install"
		return 1
	fi

	if is_running; then
		log_warn "Already running"
		return 0
	fi

	log_info "Starting LocalAI..."
	/etc/init.d/localai start
}

cmd_stop() {
	require_root
	/etc/init.d/localai stop
}

cmd_restart() {
	require_root
	/etc/init.d/localai restart
}

cmd_status() {
	load_config

	echo "=== LocalAI Status ==="
	echo ""

	if [ -x "$BINARY" ]; then
		echo "Binary: $BINARY"
		local size=$(ls -lh "$BINARY" 2>/dev/null | awk '{print $5}')
		echo "Size: $size"
	else
		echo "Binary: NOT FOUND"
		echo ""
		echo "Run: localaictl install"
		return 1
	fi

	echo ""

	if is_running; then
		echo "Service: RUNNING"
		local pid=$(pgrep -f "$BINARY" | head -1)
		echo "PID: $pid"
	else
		echo "Service: STOPPED"
	fi

	echo ""
	echo "Configuration:"
	echo "  API: http://${api_host}:${api_port}"
	echo "  Models: $models_path"
	echo "  Threads: $threads"
	echo "  Context: $context_size"

	echo ""

	# Check backend assets
	echo "Backends:"
	if [ -d "$BACKEND_ASSETS/grpc" ]; then
		local backend_count=0
		for b in "$BACKEND_ASSETS/grpc"/*; do
			[ -x "$b" ] && backend_count=$((backend_count + 1))
		done
		echo "  GRPC backends: $backend_count installed"
	else
		echo "  GRPC backends: none (using built-in)"
	fi

	echo ""

	# Check API health
	if is_running; then
		if wget -q -O /dev/null "http://127.0.0.1:$api_port/readyz" 2>/dev/null; then
			echo "API Status: HEALTHY"
		else
			echo "API Status: NOT RESPONDING (may be loading)"
		fi
	fi
}

cmd_logs() {
	if [ "$1" = "-f" ]; then
		logread -f -e localai
	else
		logread -e localai | tail -100
	fi
}

# =============================================================================
# BACKEND MANAGEMENT
# =============================================================================

cmd_backends() {
	echo "=== Available Backends ==="
	echo ""

	# Check installed backend binaries
	if [ -d "$BACKEND_ASSETS/grpc" ]; then
		echo "Installed GRPC backends:"
		for backend in "$BACKEND_ASSETS/grpc"/*; do
			[ -x "$backend" ] || continue
			local name=$(basename "$backend")
			echo "  - $name"
		done
	else
		echo "No external GRPC backends installed"
		echo "Using built-in backends from binary"
	fi

	echo ""
}

# =============================================================================
# MODEL MANAGEMENT
# =============================================================================

cmd_models() {
	load_config
	echo "=== Installed Models ==="
	echo ""

	if [ -d "$models_path" ]; then
		local count=0
		for model in "$models_path"/*.gguf "$models_path"/*.bin; do
			[ -f "$model" ] || continue
			count=$((count + 1))
			local name=$(basename "$model")
			local size=$(ls -lh "$model" | awk '{print $5}')
			echo "  $count. $name ($size)"
		done
		[ "$count" -eq 0 ] && echo "  No models installed"
	fi

	echo ""
	echo "=== Available Presets ==="
	echo "  tinyllama    - 669MB  - TinyLlama 1.1B (chat)"
	echo "  phi2         - 1.6GB  - Microsoft Phi-2 (chat)"
	echo "  mistral      - 4.1GB  - Mistral 7B Instruct (chat)"
	echo "  gte-small    -  67MB  - GTE Small (embeddings)"
	echo ""
	echo "Install: localaictl model-install <name>"
}

cmd_model_install() {
	load_config
	require_root

	local model_name="$1"
	[ -z "$model_name" ] && { echo "Usage: localaictl model-install <name|url>"; return 1; }

	mkdir -p "$models_path"

	# Preset URLs
	local url="" filename=""
	case "$model_name" in
		tinyllama)
			url="https://huggingface.co/TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF/resolve/main/tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf"
			filename="tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf"
			;;
		phi2|phi-2)
			url="https://huggingface.co/TheBloke/phi-2-GGUF/resolve/main/phi-2.Q4_K_M.gguf"
			filename="phi-2.Q4_K_M.gguf"
			;;
		mistral)
			url="https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-GGUF/resolve/main/mistral-7b-instruct-v0.2.Q4_K_M.gguf"
			filename="mistral-7b-instruct-v0.2.Q4_K_M.gguf"
			;;
		gte-small)
			url="https://huggingface.co/ggml-org/gte-small-Q8_0-GGUF/resolve/main/gte-small-q8_0.gguf"
			filename="gte-small-q8_0.gguf"
			;;
		http*)
			url="$model_name"
			filename=$(basename "$url")
			;;
		*)
			log_error "Unknown model: $model_name"
			log_error "Use preset name (tinyllama, phi2, mistral, gte-small) or full URL"
			return 1
			;;
	esac

	log_info "Downloading: $filename"
	log_info "URL: $url"
	log_info "This may take several minutes..."

	if wget -L --show-progress -O "$models_path/$filename" "$url"; then
		# Create YAML config for the model
		local model_id="${filename%.*}"

		# Embedding models need different config
		case "$model_name" in
			gte-small)
				cat > "$models_path/$model_id.yaml" << EOF
name: $model_id
backend: llama-cpp
embeddings: true
parameters:
  model: $filename
context_size: 512
threads: $threads
EOF
				;;
			*)
				cat > "$models_path/$model_id.yaml" << EOF
name: $model_id
backend: llama-cpp
parameters:
  model: $filename
context_size: $context_size
threads: $threads
EOF
				;;
		esac
		log_info "Model installed: $model_id"
		log_info "Restart service to load: /etc/init.d/localai restart"
	else
		log_error "Download failed"
		rm -f "$models_path/$filename"
		return 1
	fi
}

cmd_model_remove() {
	load_config
	require_root

	local model_name="$1"
	[ -z "$model_name" ] && { echo "Usage: localaictl model-remove <name>"; return 1; }

	local found=0
	for ext in gguf bin yaml yml; do
		if [ -f "$models_path/$model_name.$ext" ]; then
			rm -f "$models_path/$model_name.$ext"
			found=1
		fi
	done

	# Also try to match partial names
	for file in "$models_path"/*"$model_name"*; do
		[ -f "$file" ] && rm -f "$file" && found=1
	done

	[ $found -eq 1 ] && log_info "Model removed: $model_name" || log_warn "Model not found: $model_name"
}

# =============================================================================
# MAIN
# =============================================================================

case "${1:-}" in
	install) cmd_install ;;
	uninstall) cmd_uninstall ;;
	start) cmd_start ;;
	stop) cmd_stop ;;
	restart) cmd_restart ;;
	status) cmd_status ;;
	logs) shift; cmd_logs "$@" ;;
	backends) cmd_backends ;;
	models) cmd_models ;;
	model-install) shift; cmd_model_install "$@" ;;
	model-remove) shift; cmd_model_remove "$@" ;;
	help|--help|-h|'') usage ;;
	*) echo "Unknown: $1" >&2; usage >&2; exit 1 ;;
esac