Why I Wanted My Own AI Assistant

So here’s the thing. I use Claude, ChatGPT, Gemini, pretty much all of them on a daily basis. They’re great tools, genuinely. But at some point I started thinking about where all my conversations actually live. Every prompt, every response, every piece of context sits on someone else’s servers. I can’t move it. I can’t switch providers mid-conversation. I can’t plug it into my own workflows the way I want.

For casual use that’s totally fine. But I was building projects that needed AI capabilities integrated into my own infrastructure, and relying entirely on third-party web apps didn’t feel right anymore. I wanted something I could run myself, on my own servers, with full control over the models and the data.

That’s when I found OpenClaw.

What OpenClaw Actually Is

OpenClaw is a self-hosted AI assistant platform. You install it on your own server, connect it to whatever AI model you want, and interact with it through messaging apps you already use. WhatsApp, Telegram, Slack, Discord, and a bunch of others. So instead of going to a web app, you just chat with your AI assistant directly from your phone or your team’s Slack workspace.

The project has an interesting history too. It went through several name changes, from Warelay to Clawd to Molty, before finally landing on OpenClaw. The rename happened after Anthropic reached out about trademark concerns with “Clawd”, and the team migrated everything in about three hours. Different name, same project. If you’re curious about the full story, there’s the official lore which is genuinely entertaining.

There are also lighter community forks like NanoClaw, PicoClaw, ZeroClaw, and many others.

Deploying It on AWS

For this deployment, I used an EC2 instance instead of Lightsail because I wanted the flexibility to manage my own security groups, networking, and instance types. I went with a t4g.medium which is ARM-based and gives you a good balance of performance and cost for running something like OpenClaw.

One thing I want to mention here is AWS Spot Instances. They’re spare compute capacity that AWS sells at a steep discount, sometimes up to 90% cheaper than On-Demand pricing. The tradeoff is that AWS can reclaim the instance with a two-minute warning if they need the capacity back. For something like an AI assistant that you can quickly redeploy, Spot Instances are perfect. I use them in my setup and the savings are significant.

Setting Up the Server

You start by creating an EC2 instance. You can just use the default VPC that AWS provides, no need to set up a custom one unless you have specific networking requirements. Make sure the security group allows SSH on port 22 and web traffic on port 8080. If you want a static IP, attach an Elastic IP to the instance so the address doesn’t change when you reboot.

Installing OpenClaw

Once I SSH’d in, OpenClaw has a one-liner installer that handles everything:

curl -fsSL https://openclaw.ai/install.sh | bash

If you prefer to configure things manually, you can skip the interactive onboarding:

curl -fsSL https://openclaw.ai/install.sh | bash -s -- --no-onboard

Choosing an AI Provider

OpenClaw doesn’t ship with its own model, so you connect it to whatever provider you prefer. In my case I went with OpenRouter because it lets me swap between Claude 4.6, GPT-5, Gemini, and others with just a config change. If you’re already deep in the AWS ecosystem, Bedrock is another solid option. You can also plug in API keys directly from OpenAI, Anthropic, or Google.

openclaw config set env.OPENROUTER_API_KEY 'your-api-key'
openclaw config set agents.defaults.model.primary 'openrouter/anthropic/claude-sonnet-4'

Getting It Online

By default OpenClaw only listens on localhost, so you need to open it up:

openclaw config set gateway.bind lan

I also set up Telegram integration since I wanted to chat with the assistant from my phone. You grab a bot token from @BotFather and add it to the config:

openclaw config set channels.telegram.enabled true
openclaw config set channels.telegram.botToken 'your-bot-token'

Then restart the gateway and you’re done:

openclaw gateway restart

Your instance should be live at http://your-instance-ip:8080.

Automating Everything

The steps above are straightforward, but I also packaged the entire deployment into a repository that automates all of it. Infrastructure provisioning, OpenClaw installation, configuration, Spot Instance setup, everything runs through a single deploy command.

Why I Think This Is Worth Doing

Running your own AI assistant gives you something the hosted platforms can’t: complete ownership. You pick the models, you decide where the data lives, you control who has access. When I want to experiment with a different LLM, I change one line in a config file instead of signing up for another service.

It’s not for everyone, and the hosted platforms are excellent for what they are. But if you’re the kind of person who likes to own your infrastructure and understand exactly how things work under the hood, this is a really satisfying project.

The Repository

I’ve open-sourced my complete setup. Everything you need to deploy OpenClaw on AWS is in here.

Clone it, deploy it, tweak it for your own needs. The repo includes the full infrastructure code, automated setup scripts, example configs, and a Makefile for common operations.