How They Work
What are 'they' made of?
A plain-English explanation of the components, security layers, and design decisions that make AI assistants reliable and safe.
Core Components
1. The Gateway
This is the brain. It runs on your hardware (or cloud server) and manages everything: reading messages, calling tools, talking to AI providers, and sending responses.
Key point: The gateway never runs on your personal computer. It lives on a separate device or server so it can work 24/7 without affecting your daily machine.
Hosting costs: $10-50/month (Raspberry Pi, VPS, or dedicated server).
2. Message Channels
How you talk to your assistant. Telegram, Discord, Slack, WhatsApp, or even email. The channel is just the interface—the gateway does the heavy lifting.
Why it matters: You can switch channels without rebuilding your assistant. The channel and the logic are completely separate.
3. AI Provider
OpenAI, Anthropic, Google, or others. This is where the actual language model lives. Your assistant sends prompts, gets responses, and decides what to do next.
Cost note: You pay the AI provider directly based on usage. No markup from us. Typical costs: $20-200/month depending on usage and which models you use.
4. Tools and Skills
What your assistant can actually do. Read email, check calendars, pull weather forecasts, search the web, send notifications. Each "skill" is a small script that extends capabilities.
Security layer: Tools run in isolation. If one breaks, it doesn't crash the whole system. And you control which tools have access to what.
5. Memory and Context
Your assistant remembers conversations, preferences, and history. This is stored locally on your gateway—not in the cloud (unless you choose otherwise).
Privacy win: Your data stays on your hardware. AI providers see prompts and responses but don't get your entire conversation history.
Security Layers
Defense in Depth
No single security measure is perfect. So we stack multiple layers:
- 1.Isolation: Your assistant runs on a separate device/server—not your laptop.
- 2.Separate email identity: Your assistant sends from its own address, never impersonates you.
- 3.Read-only by default: Start with read permissions only. Add write access gradually as you build trust.
- 4.Tool sandboxing: Each skill runs in isolation. If one fails, the rest keep working.
- 5.Credential management: API keys and passwords stored in encrypted vaults (1Password, Bitwarden).
The unsolved problem: Prompt injection (tricking the AI via crafted input) is still an active research area. We defend by limiting blast radius—if someone tricks your assistant, they can't escalate to your personal email or financial accounts because those are isolated.
Key Design Decisions
Why separate hardware?
Your assistant needs to run 24/7. If it's on your laptop, it stops working when you close the lid. A Raspberry Pi, old laptop, or cloud server keeps it always-on without tying up your daily machine.
Why not use cloud-hosted AI assistants?
You could. But then your data lives in someone else's servers, and you're locked into their pricing and features. Self-hosting gives you control: swap AI providers, add custom skills, keep data local, and pay only for what you use.
Why multiple platforms (OpenClaw, NanoClaw, PicoClaw)?
Different trade-offs. OpenClaw is feature-rich and mature. PicoClaw is ultra-lightweight for low-power hardware. NanoClaw prioritizes security with minimal code surface. We help you pick based on your needs, not ours.
What about reliability and uptime?
Single-server setups are fine for most people. If you need 99.9% uptime, we can set up redundancy: multiple gateways, failover, health checks. But that adds complexity and cost—only worth it if downtime truly hurts your business.
Platform Comparison
| Platform | Best For | Min Hardware | Community |
|---|---|---|---|
| OpenClaw | Most users, richest feature set | 1GB RAM, any OS | Largest, most active |
| PicoClaw | Ultra-low-power setups | 10MB RAM, tiny devices | Growing, hardware-focused |
| NanoClaw | Security-first, minimal code | 512MB RAM, Linux | Small, security-minded |
Need help choosing? Compare platforms in detail →
Ready to Build Your Assistant?
We handle the technical details—you get a working, secure AI assistant tailored to your workflow.