I Built an AI System That Manages My Entire Fitness Life. Here's How.

The Problem With Fitness Data

I've worn three sleep trackers to bed - Garmin, WHOOP, and 8Sleep. They all tell me something different every night.

That's not a complaint. They're measuring different things. But the problem is none of them talk to each other. Garmin shows my sleep differently than WHOOP. 8Sleep has its own score. WHOOP gives me recovery. Garmin gives me training status. None of it connects.

And the strength training side? Speediance and Tonal are walled gardens. They don't export data easily. I was stuck manually tracking everything in spreadsheets - the exact thing I bought these machines to avoid.

So I built a system to fix it.

What I'm Running

I'm running OpenClaw on an M3 Ultra Mac. ARIA is my main assistant, handling the dashboard, task board, and coordinating data from multiple connectors. I use a mix of local and hosted models for cost-aware operation.

The Morning Report

Every morning, the system runs and pulls data from everywhere:

• WHOOP - recovery score, strain, sleep performance
• Garmin - sleep analysis, training status, HRV
• 8Sleep - sleep score, time asleep, HRV
• Speediance - yesterday's lifting volume, workout details
• Cronometer - nutrition, protein, calories

Then it generates a morning report. Not just data - analysis. It tells me:

• How recovered I am
• How hard I trained yesterday
• What my training status should be today
• Whether I should push or rest

This is the insight I couldn't get from any single app. None of them know about the others. None of them can correlate sleep + lifting + nutrition together.

The Nightly Report

Every night, another report runs:

• Did I hit my protein target?
• How much did I lift today?
• What's my cumulative strain for the week?
• How's my recovery trending?

It's like having a coach who never sleeps checking in every day.

The Connectors

OpenClaw didn't have native connectors for most of this. I built them.

Actually - I told OpenClaw what I wanted. It built the connectors.

For Speediance, there's an unofficial API endpoint that pulls workout data. I fed that into OpenClaw. Now it pulls my lifting volume automatically every day.

For Garmin and WHOOP, I'm using their APIs. The data flows into the system, gets normalized, and becomes useful.

For 8Sleep, it's trickier. The data export is inconsistent. Some nights work, some don't. That's on my list to fix.

Why Mac Mini?

I tried running this on a Windows PC first. It was hot garbage. Setup was a nightmare. Could barely do anything.

Then I tried it on an M3 Ultra Mac. Night and day difference. Everything worked out of the box. OpenClaw runs smoothly. The API calls are fast. It's reliable.

The lesson: don't cheap out on the hardware. If you want this to work, use a capable Mac.

The Free Providers

I'm using several free LLM providers through OpenClaw:

• Google - solid, but rate limits hit fast
• Nvidia - Kimmy 2.5 is excellent when it works, but can hang
• MiniMax - my daily driver. $10/month for the cheapest plan. Reliable, slow, but deliberate. The slowness is actually a benefit - it doesn't burn tokens quickly.
• Mistral - use occasionally, haven't hit limits

For heavy reasoning tasks, I switch to hosted models like Claude (Anthropic) on demand. The cost adds up quickly, so I reserve that for complex multi-step tasks where local models struggle.

The Agent Swarm

Here's the pattern I use:

When I have a complex task, I spawn multiple agents in parallel. MiniMax coordinates them. Each agent handles a piece of the task - one pulls data, another formats it, a third checks for errors. Then MiniMax puts it all together.

This saves tokens. Running everything through premium hosted models would burn through my budget quickly. By distributing work to free providers and using hosted models only for assembly, I keep costs manageable.

The Fitness Dashboard

I built a custom dashboard using OpenClaw. It shows:

• All my daily metrics in one view
• Trend graphs for sleep, recovery, lifting
• Weekly and monthly summaries
• Training status recommendations

It's hosted on GitHub Pages. I can access it from anywhere. So can anyone else - I'm open about my training data. If people want to see how I'm doing, they can.

The BJJ Tracker App

One of the projects I'm building is a BJJ technique tracker. The vision:

• 150+ techniques indexed
• Filter by position, belt level, submission type
• Track which ones I've trained, favorites, gaps
• Video links for every technique

OpenClaw built the initial structure. Then I switched to Codex for the frontend edits. That's the workflow: OpenClaw for architecture and logic, Codex for UI tweaks. Minor edits aren't worth the context switching with an LLM.

What I've Learned

Free models work. MiniMax at $10/month handles 90% of what I need. Claude only for the hard stuff.

Hardware matters. A capable Mac (M3 Ultra in my case) is the sweet spot. Reliable, powerful enough.

Agents are additive. Running one agent is helpful. Running three in parallel, coordinated, is transformative.

Fitness data should be unified. No single app knows everything. Building the bridge between Garmin + WHOOP + Speediance + 8Sleep gave me insights I couldn't get anywhere else.

This is as big as the internet. I've been around since the early web. I've seen transformations before. OpenClaw - and AI agents more broadly - are the most transformative tech I've encountered. The iPhone was incredible. This is bigger.

What's Next

I'm continuing to iterate on the dashboard. The second brain is growing - reminders, automations, integrations.

The BJJ tracker needs work. It's bare bones right now. But the foundation is there.

I'm also looking at home automation integration. Alexa, smart lights, climate control - having an agent that knows my training load and adjusts my environment accordingly.

If you want to see the fitness reports, they're live at my GitHub Pages. Morning and nightly, running automatically. That's the power of this setup.

The Bottom Line

I didn't build this because the apps weren't good enough. I built it because no single app could see the whole picture. Sleep + lifting + nutrition + recovery - they're all connected. The data should be too.

OpenClaw made it possible. Not easy - but possible. And once it's running, it runs itself.