Self-Hosted AI Assistant: Complete Channel Guide

A self-hosted AI assistant lives on your server, responds on your messaging apps, and keeps your data private. No monthly SaaS fees eating into your budget, no rate limits during peak usage, no vendor lock-in. This guide covers all the options for building your own.

Why self-host your AI assistant?

• Privacy - Conversations stay on your infrastructure
• Cost control - Pay for API usage, not per-seat SaaS fees
• Customization - System prompts, tools, integrations you control
• Multi-channel - One assistant across all your messaging apps
• No limits - No artificial rate limits or usage caps

Architecture overview

A self-hosted AI assistant has three main components:

1. Gateway - Connects to messaging platforms (WhatsApp, Discord, etc.)
2. AI Backend - Processes messages with Claude, GPT, or local models
3. State Store - Persists conversation history and user preferences

You can build each component yourself, or use something like OpenClaw which bundles them together.

Channel comparison

Quick start guides by channel

Telegram (easiest)

Get a bot token from @BotFather in 30 seconds, write 50 lines of code, done.

→ Full Telegram AI bot guide

Discord

Create app in Discord Developer Portal, enable intents, deploy bot. Great for community servers.

→ Full Discord AI bot guide

WhatsApp

Uses Baileys library to connect via WhatsApp Web protocol. Requires QR code pairing.

→ Full WhatsApp AI bot guide

Slack

Create Slack app, configure OAuth scopes, deploy with Bolt SDK. Perfect for team assistants.

→ Full Slack AI bot guide

iMessage

Requires a Mac running BlueBubbles. Full iMessage features including reactions and effects.

→ Full iMessage AI bot guide

Signal

Uses signal-cli for end-to-end encrypted AI conversations. Privacy-first option.

→ Full Signal AI bot guide

AI model options

Cloud APIs (recommended for most)

• Claude (Anthropic) - Best for conversation, coding, analysis
• GPT-4 (OpenAI) - Versatile, good ecosystem
• Gemini (Google) - Good for Google Workspace integration

Self-hosted models

• Llama 3 - Meta's open model, runs on consumer GPUs
• Mistral - Efficient, good quality/size ratio
• Qwen - Strong multilingual support

Self-hosted models need significant hardware (16GB+ VRAM for good results) but eliminate API costs and latency.

Infrastructure requirements

Minimum (cloud APIs)

• VPS with 1 CPU, 1GB RAM ($5/month)
• Or Raspberry Pi for home hosting
• API keys for your chosen model

For self-hosted models

• GPU with 16GB+ VRAM (RTX 3090, 4090, or Mac with M1/M2)
• 32GB+ system RAM
• Fast SSD for model weights

For iMessage

• Mac Mini or similar always-on Mac
• macOS with iMessage configured
• BlueBubbles server app

Cost breakdown

OpenClaw: The all-in-one option

OpenClaw (formerly MoltBot/Clawdbot) bundles everything:

• Multi-channel gateway - WhatsApp, Telegram, Discord, Slack, iMessage, Signal, and more
• Model switching - Claude, GPT-4, Gemini, or local models
• Conversation persistence - Context across sessions
• Web UI - Configure without touching code
• Bring your own key - Use existing Claude/ChatGPT subscriptions

Deploy on Molted (managed) or self-host on your own infrastructure.

Security considerations

API key management

• Never commit keys to git
• Use environment variables or secrets manager
• Rotate keys periodically
• Set spending limits on API accounts

Access control

• Allowlist specific users/numbers
• Rate limit per user
• Log all interactions for audit
• Block prompt injection attempts

Data privacy

• Encrypt conversation storage at rest
• Auto-delete old conversations
• Be aware of what data flows to AI providers
• Consider local models for sensitive use cases

Monitoring and reliability

# Basic health check endpoint
app.get('/health', (req, res) => {
  res.json({
    status: 'ok',
    channels: {
      telegram: telegramConnected,
      discord: discordConnected,
      whatsapp: whatsappConnected,
    },
    uptime: process.uptime(),
  });
});

# Monitor with Uptime Kuma, Healthchecks.io, or similar

Getting started

The fastest path to a working self-hosted AI assistant:

1. Pick one channel - Start with Telegram (easiest) or the platform you use most
2. Choose your model - Claude or GPT-4 for best quality, local for privacy
3. Deploy - VPS, home server, or Molted
4. Add more channels - Once the first works, expand

Or skip the setup entirely and deploy OpenClaw on Molted. You get all channels configured in minutes.

Related guides

• Mac Mini AI Server - Home server setup for 24/7 AI
• OpenClaw Pricing Explained - Avoid the $700/month horror stories
• Claude vs ChatGPT - Which model for your assistant?