OpenClaw: One AI, All Your Chat Apps
I use WhatsApp for family, Telegram for work groups, and Discord for hobby stuff. Each one has its own set of bots and integrations. It gets messy fast. OpenClaw solves this by connecting all of them to a single AI assistant running on my own server.
The setup took about an hour. First I installed OpenClaw globally with npm. Then I ran the onboard command, which walks you through connecting each channel. WhatsApp required scanning a QR code. Telegram needed a bot token from BotFather. Discord was straightforward with a standard bot invite.
What’s interesting about OpenClaw is that it keeps a single conversation context across channels. If I start talking to the assistant on WhatsApp and then continue on Discord, it remembers what we were discussing. The sessions are tied to my identity, not to a specific app.
The multi-agent routing is where this gets useful. I configured it to use different AI models for different channels. WhatsApp gets a lighter model for quick responses. Slack gets a more capable one for work-related questions. The routing is defined in a JSON config file. It’s simple but flexible.
Tool execution is available through the assistant. I set up filesystem access to a specific project directory and limited shell commands to git and npm. The permissions are explicit. The assistant can’t just do anything on my machine. You define what it can access and what commands it’s allowed to run.
Privacy is the main selling point. Everything runs on your own infrastructure. Messages don’t go through third-party servers. There’s no telemetry. The assistant processes your messages locally and sends responses back through whatever channel you’re using.
The mobile companion apps add some device-specific capabilities. Camera access, location sharing, push notifications. They’re optional but useful if you want the assistant to interact with your phone’s hardware.
I’ve been running it for a few weeks now. The quality depends on which model you connect it to. Using a local model, responses are slower but completely private. Using a cloud API through OpenClaw’s gateway, responses are fast but your messages leave your machine. You get to choose.
The webchat interface works fine for quick testing. It’s a simple browser page you can access from any device on your network. Not meant to replace your actual chat apps, just good for when you want to experiment without configuring everything.
Setup isn’t completely painless. The configuration files can be a bit opaque if you’re not familiar with JSON. The documentation covers the basics but leaves some edge cases undocumented. You’ll spend time reading code and trial-and-error when something doesn’t work as expected.
Still, it’s a solid approach to the problem of fragmented AI access. One assistant, multiple channels, all controlled by you.
OpenClaw does one thing and does it well. If you want a single AI assistant across all your chat apps without handing your data to a cloud provider, it’s worth trying.