One-click installer for a local LLM stack (Gemma 4 + llama.cpp + OpenClaw)

openclaw_gemma4_install.sh

#!/usr/bin/env bash

: <<'DOC'
🚀 One-click installer for a local LLM stack (Gemma 4 + llama.cpp + OpenClaw)

What this script does:
- Installs required dependencies (git, cmake, build tools)
- Clones and builds llama.cpp
- Runs a local LLM server using Gemma 4 GGUF model
- Installs OpenClaw (OpenAI-compatible API wrapper)
- Connects OpenClaw to the local model
- Cleans up background processes on exit

Usage:
  chmod +x openclaw_gemma4_install.sh
  ./openclaw_gemma4_install.sh

Notes:
- Runs server on port 8080
- Requires Linux (Ubuntu/Debian recommended)
- Ensure enough RAM/CPU for model
DOC

# Exit immediately if any command fails
set -e

echo "🚀 Installing local LLM stack..."

# --------------------------------------------------
# 1. Install required system dependencies
# --------------------------------------------------

# Check if git is installed; install if missing
if ! command -v git &> /dev/null; then
  echo "Installing git..."
  sudo apt-get update && sudo apt-get install -y git
fi

# Check if cmake is installed; install build tools if missing
if ! command -v cmake &> /dev/null; then
  echo "Installing build tools (cmake, gcc, etc.)..."
  sudo apt-get install -y build-essential cmake
fi

# --------------------------------------------------
# 2. Clone llama.cpp (if not already present)
# --------------------------------------------------

# Clone repository only if it doesn't exist locally
if [ ! -d "llama.cpp" ]; then
  echo "Cloning llama.cpp repository..."
  git clone https://github.com/ggerganov/llama.cpp.git
fi

# Move into llama.cpp directory
cd llama.cpp

# Build llama.cpp using CMake (Release mode)
echo "Building llama.cpp..."
cmake -B build
cmake --build build --config Release

# --------------------------------------------------
# 3. Start local LLM server (llama-server)
# --------------------------------------------------

echo "Starting llama-server..."

# Run llama-server in background
./build/bin/llama-server \
  -hf ggml-org/gemma-4-26b-a4b-it-GGUF:Q4_K_M \   # HuggingFace GGUF model
  --port 8080 \                                  # API port
  --host 0.0.0.0 &                               # Bind to all interfaces

# Store process ID for cleanup
SERVER_PID=$!

# Wait for server to initialize (simple delay)
sleep 10

# --------------------------------------------------
# 4. Install OpenClaw (OpenAI-compatible interface)
# --------------------------------------------------

cd ..

# Install OpenClaw if not already available
if ! command -v openclaw &> /dev/null; then
  echo "Installing OpenClaw..."
  pip install openclaw
fi

# --------------------------------------------------
# 5. Configure & run OpenClaw
# --------------------------------------------------

echo "Starting OpenClaw..."

# Connect OpenClaw to local llama.cpp server
openclaw onboard --non-interactive \
  --auth-choice custom-api-key \                      # Use custom API key mode
  --custom-base-url "http://127.0.0.1:8080/v1" \     # Local LLM endpoint
  --custom-model-id "ggml-org-gemma-4-26b-a4b-gguf" \# Model identifier
  --custom-api-key "llama.cpp" \                     # Dummy API key
  --secret-input-mode plaintext \                    # No hidden input
  --custom-compatibility openai \                    # OpenAI-compatible API
  --accept-risk                                     # Skip confirmations

# --------------------------------------------------
# Cleanup: ensure server stops when script exits
# --------------------------------------------------

# Kill llama-server when script exits (CTRL+C or failure)
trap "echo 'Stopping llama-server...'; kill $SERVER_PID" EXIT