Streaming My Vitals To "Dr. Claw"

It beats paying my deductible...

These are just a few snippets from my new AI Doctor. Okay, it’s not really a doctor but it was fun to setup nonetheless.

I created a new openclaw agent that is managed through discord in a specific channel, that has near live access to my heart rate, hrv, sleep data, and more. This is all data coming from a Helio Strap.

This company has easy to reverse engineer products, so you aren’t tied to their mobile app or cloud platform for data access.

I setup Gadgetbridge next, as I run Graphene. (p.s. I am super excited to see Motorola as a partner next year)

Gadgetbridge lets you automatically export a sqlite database on a schedule with all collected data. You can already see where this is going. I set this up to auto export every few hours, and a termux command to sync from that exported database to my openclaw server. This server uses Tailscale so an rsync / scp doesn’t require manually setting up any ssh keys.

The sync drops the file right into the agent’s workspace.

At this point I ran openclaw agents add health and then bound it to a discord channel. Funny enough the agent sometimes gets confused that health = health checks. Probably better to name your agent “doctor” but then it might think you’re asking about openclaw doctor LOL.

Anyways, with that in place, I tell the agent “You are my doctor, you have full access to my vitals in the sqlite database Gadgetbridge.db in your workspace directory. View the schema and then persist it to your soul and to tools.md with all the table information and your purpose”

I did setup a few more things, like asking it to give me a daily report at a set time, weekly reports, and to alert me anytime values look off or I look unhealthy.

It’s really just a for fun experiment in using openclaw’s built in tooling and playing around. I tend to use Ollama Cloud for model access. I already have Claude Enterprise and Codex at work, so it keeps me up to date on how the open source models are progressing, by using them in Openclaw.

I have so many providers, I setup LiteLLM so opencode can reference a model name and I can swap it out at the LiteLLM level instead of modifying opencode on the 5 different places I have it installed. 😂

I also give read only access to a lot of other development tools, so I can have some quick help when I’m on the go, vs dropping into an ssh session and claude code from my phone. I know Claude also has a built in way to do that now too.

I like using qwen 3.5 397b and GLM-5 through ollama cloud currently for these types of agents.

Random Thoughts

You can get a nice, safe openclaw setup if you only use a couple skills that you vet yourself, along with using external services in read-only mode.

I am able to work on git repos, but the agent cannot commit for me. I need to ssh in, and ssh forwarding allows the container to request signing via 1password’s cli. The container never sees the private key. I may post more about this later.