Prima di iniziare: Comprendere cosa stai costruendo
Most AI installation guides treat software as a black box: download, install, forget. clawbot is different because you're not installing an app—you're establishing infrastructure . Think of this as setting up your own private email server, except instead of managing messages, you're orchestrating an AI that can genuinely control your digital environment.
This guide explains what each step does and why it matters. Understanding the architecture makes troubleshooting intuitive and customization straightforward.
What You'll Have After 30 Minutes
- A Gateway service running persistently on your device, coordinating all AI activity
- Connection to your chosen AI model (Claude, GPT-4, or free local Ollama)
- At least one messaging channel (WhatsApp, Telegram, etc.) linked to your AI
- A functioning automation that demonstrates genuine system control
- Understanding of how to extend, customize, and troubleshoot your setup
Prerequisiti: Cosa ti serve realmente
Prima di scaricare qualsiasi cosa, verifica che il tuo sistema soddisfi questi requisiti. Ognuno di essi serve uno scopo specifico:
Requisiti hardware
- Operating System: macOS 11+, Linux (Ubuntu 20.04+, Debian 11+), or Windows 10+ with WSL2
- RAM: Minimum 2GB available (4GB recommended for smoother operation)
- Disk Space: 500MB for core installation plus space for AI model caching
- Network: Stable internet connection for AI API calls (unless using local Ollama)
Dipendenze software
clawbot requires Node.js 22 or higher. Why this specific version? Newer Node.js versions include critical performance improvements and security patches that clawbot's real-time communication relies on. The installer typically handles this, but manual installations require it pre-installed.
Accesso ai modelli AI (scegli uno)
Cost-Benefit Analysis: Which AI Model Should You Choose?
Option 1: Anthropic Claude (Recommended for most users)
- Cost: ~$20-40/month for typical personal use
- Advantage: Exceptional reasoning, strong security awareness, long context windows
- Get API key: console.anthropic.com
Option 2: OpenAI GPT-4
- Cost: ~$25-50/month depending on usage patterns
- Advantage: Broad knowledge base, excellent for creative tasks
- Get API key: platform.openai.com
Option 3: Local Ollama Models (Completely free)
- Cost: $0 (runs entirely on your hardware)
- Advantage: Zero ongoing costs, complete privacy, offline operation
- Disadvantage: Slower responses, requires more powerful hardware (8GB+ RAM recommended)
- Setup: ollama.ai
Passo 1: Scarica il pacchetto giusto per la tua piattaforma
clawbot distributes platform-specific installers that bundle everything needed. Why multiple formats? Different operating systems handle background services differently—the installer configures these correctly.
Installazione macOS
Scarica l'installer DMG per l'esperienza più semplice:
curl -fsSL https://github.com/steipete/clawbot/releases/download/v2026.1.23/clawbot-2026.1.23.dmg -o clawbot.dmg
Or download directly from GitHub Releases .
What the installer does:
-
Installs the Gateway service to
/Applications/clawbot.app -
Creates configuration directory at
~/.clawbot/ - Configures system permissions for automation capabilities
- Optionally sets up launch-at-startup (recommended)
Installazione Linux
Installa tramite npm (consigliato) o Docker:
npm install -g clawbot@latest
Quindi inizializza la configurazione:
clawbot init
Setting up as a systemd service (optional but recommended):
clawbot install-service
Ciò garantisce che clawbot si avvii automaticamente all'avvio e si riavvii in caso di crash.
Installazione Windows (WSL2)
Windows richiede WSL2 (Windows Subsystem for Linux). Se non installato:
wsl --install
Quindi segui le istruzioni per Linux all'interno del tuo ambiente Ubuntu WSL2.
⚠️ Windows Native Support
Il supporto nativo per Windows (senza WSL2) è sperimentale. WSL2 offre una migliore compatibilità con le integrazioni di sistema di Node.js e l'esecuzione di comandi shell.
Deployment Docker
Per compatibilità multipiattaforma o deployment su server:
docker pull ghcr.io/steipete/clawbot:latest
Esegui con configurazione persistente:
docker run -d \
--name clawbot \
-v ~/.clawbot:/root/.clawbot \
-p 18789:18789 \
ghcr.io/steipete/clawbot:latest
Understanding the Docker flags:
-
-d: Runs container in background (daemon mode) -
-v ~/.clawbot:/root/.clawbot: Persists configuration between restarts -
-p 18789:18789: Exposes Gateway WebSocket port for local clients
Passo 2: Installa e avvia il Gateway
The Gateway is clawbot's core—a persistent service that manages all communication between messaging platforms, AI models, and your system. Think of it as the central nervous system : channels send messages to the Gateway, which routes them to the appropriate AI model, receives responses, and delivers them back.
Esegui l'installer
macOS: Open the downloaded DMG and drag clawbot to Applications. Launch it.
Linux/WSL: After npm installation, start the Gateway:
clawbot gateway start
Verifica che sia in esecuzione:
clawbot gateway status
You should see:
Gateway is running on ws://127.0.0.1:18789
🔍 What's Happening Behind the Scenes
Il Gateway crea un server WebSocket sulla porta 18789 (personalizzabile). Questo server, accessibile solo localmente, accetta connessioni da:
- Channel plugins (WhatsApp, Telegram, etc.) that forward user messages
- Node clients (mobile apps, web interfaces) for direct interaction
- Skills and tools that extend functionality
Tutta la comunicazione rimane sulla tua macchina a meno che tu non configuri esplicitamente l'accesso remoto (trattato nelle guide avanzate).
Passo 3: Collega il tuo cervello AI
Con il Gateway in esecuzione, necessita di un modello AI per generare risposte. Questo passaggio configura quali modelli utilizzerai e autentica l'accesso API.
Configura il provider AI
Apri il file di configurazione:
clawbot config edit
Or manually edit
~/.clawbot/clawbot.json
. Add your API credentials:
{
"aiProviders": {
"anthropic": {
"apiKey": "sk-ant-your-api-key-here",
"model": "claude-3-5-sonnet-20250219",
"enabled": true
}
},
"defaultProvider": "anthropic"
}
For local Ollama instead:
{
"aiProviders": {
"ollama": {
"baseURL": "http://localhost:11434",
"model": "llama3.1:8b",
"enabled": true
}
},
"defaultProvider": "ollama"
}
Security Best Practice: API Key Management
Never commit
clawbot.json
to version control if it contains API keys. Consider using
environment variables for production deployments:
export ANTHROPIC_API_KEY="your-key-here"
clawbot legge automaticamente dalle variabili d'ambiente se presenti.
Passo 4: Collega i tuoi canali di comunicazione
Channels are how you interact with your AI—WhatsApp for mobile, Telegram for desktop, Discord for team collaboration. Each channel connects independently to the Gateway, and clawbot maintains a unified conversation history across all of them .
Collega il tuo primo canale: Esempio WhatsApp
Installa il plugin del canale WhatsApp:
clawbot channel add whatsapp
Questo avvia una configurazione interattiva:
- A QR code appears in your terminal
- Open WhatsApp on your phone → Settings → Linked Devices → Link a Device
- Scan the QR code
- clawbot confirms connection and saves session credentials
How it works: clawbot uses the WhatsApp Web protocol (via Baileys library) to establish a persistent connection. Your phone remains the "primary" device; clawbot acts as a linked secondary device—just like WhatsApp Web in your browser.
⚠️ WhatsApp Account Safety
WhatsApp's terms of service prohibit bot usage. While clawbot implements rate limiting and human-like behavior patterns, we recommend using a secondary phone number rather than your primary personal WhatsApp account. Consider:
- Getting a Google Voice or similar virtual number
- Using an old SIM card you don't actively use
- Trying Telegram instead (officially supports bots)
Canali alternativi
Ogni canale ha requisiti di configurazione unici. Opzioni popolari:
Telegram (Recommended for beginners):
clawbot channel add telegram
Segui il prompt per autenticarti con il tuo numero di telefono. Telegram supporta ufficialmente i bot, rendendola l'opzione più sicura.
Discord (Great for teams):
clawbot channel add discord
Requires creating a Discord bot application at discord.com/developers . Copy the bot token when prompted.
Full channel documentation: docs.clawd.bot/channels
Passo 5: Esegui il tuo primo task autonomo
Con tutto configurato, verifichiamo che clawbot possa effettivamente controllare il tuo sistema, non solo rispondere con testo.
Testa l'integrazione di sistema
Apri il tuo canale di messaggistica collegato (WhatsApp, Telegram, ecc.) e invia questo messaggio:
"Create a text file called test.txt in my Documents folder with today's date and a motivational quote."
What should happen:
- clawbot receives your message through the channel
- Gateway routes it to your configured AI model
- AI generates a response plan including shell commands
-
clawbot executes:
echo "..." > ~/Documents/test.txt - You receive confirmation: "Created test.txt with..."
- Check your Documents folder—the file exists
🎉 You Just Demonstrated Genuine AI Automation
What distinguishes clawbot from ChatGPT or Siri: it didn't just tell you how to create the file—it actually created it . This same capability extends to:
- Managing your calendar and sending meeting invites
- Monitoring log files and alerting you to errors
- Pulling data from APIs and generating reports
- Controlling smart home devices via Home Assistant
- Running development workflows (tests, builds, deployments)
Cosa costruire dopo
La tua infrastruttura AI è operativa. Ecco come espandere le sue capacità:
1. Installa Skill da ClawdHub
Le Skill sono moduli di automazione predefiniti. Sfoglia il marketplace:
clawbot skills browse
Skill di avvio popolari:
- google-calendar : Natural language calendar management
- home-assistant : Smart home control integration
- github-assistant : Repository management and PR reviews
- daily-briefing : Morning summary of calendar, weather, news
2. Crea automazione personalizzata
Skills are just folders with a
SKILL.md
file. Create your own:
mkdir -p ~/.clawbot/skills/my-automation
echo "# My Custom Skill" > ~/.clawbot/skills/my-automation/SKILL.md
Full guide: docs.clawd.bot/skills/creating
3. Configura comportamenti proattivi
Unlike traditional AI, clawbot can initiate contact. Enable heartbeat monitoring:
clawbot config set heartbeat.enabled true
clawbot config set heartbeat.interval "every day at 8am"
Ora clawbot ti invierà proattivamente un riepilogo mattutino senza che tu lo chieda.
4. Esegui il deployment nel cloud (Opzionale)
Eseguire clawbot sul tuo laptop funziona, ma il deployment nel cloud garantisce disponibilità 24/7. Guide disponibili per:
- DigitalOcean Droplets ($6/month)
- AWS EC2 Free Tier (12 months free)
- Hetzner Cloud (€4/month)
- Home Raspberry Pi (one-time hardware cost)
Problemi comuni e soluzioni
Gateway Won't Start
Symptom:
Error: EADDRINUSE: address already in use
Cause: Another process is using port 18789.
Solution: Find and kill the conflicting process:
lsof -ti:18789 | xargs kill -9
Or configure clawbot to use a different port in
clawbot.json
.
AI Model Returns Errors
Symptom: "API authentication failed" or "Invalid model specified"
Cause: Incorrect API key or model name in configuration.
Solution: Verify your API key is valid:
curl https://api.anthropic.com/v1/messages \
-H "x-api-key: YOUR_KEY" \
-H "anthropic-version: 2023-06-01"
Controlla i nomi dei modelli nella documentazione del provider (Anthropic, OpenAI).
WhatsApp Disconnects Frequently
Symptom: QR code authentication required every few days.
Cause: Session credentials not persisting correctly.
Solution:
Ensure
~/.clawbot/channels/whatsapp/
has write permissions:
chmod 700 ~/.clawbot/channels/whatsapp/
Verifica anche che il tuo telefono rimanga connesso a Internet: WhatsApp richiede che il dispositivo primario sia online.
Ora stai eseguendo la tua infrastruttura AI
What you've built isn't just a chatbot—it's a personal automation platform that combines the intelligence of frontier AI models with the capability of local system control. Unlike cloud services that constrain what AI can do, your clawbot installation has the same access to your digital environment that you do.
Questo potere comporta responsabilità: clawbot può eseguire qualsiasi comando tu approvi. Il sistema di sandboxing e permessi protegge da danni accidentali, ma in definitiva stai concedendo a un'AI un controllo significativo. Inizia con automazioni semplici, comprendi come vengono prese le decisioni e amplia gradualmente le capacità man mano che costruisci fiducia.
Pronto per approfondire?
Esplora configurazioni avanzate, sviluppo di skill e approfondimenti architetturali