Build your first Loop

A walk-through from an empty workspace to a deployed Health Loop you’ve triggered once. Should take under five minutes.

1. Create the loop

From your loops list, click New Health Loop.

You’ll land on the editor with a starter HealthLoop class pre-filled. Give the loop a name — slugs work well, e.g. daily-cardio-review — and replace the body with whatever you want it to do. Here’s the smallest useful shape:

export class HealthLoop extends Loop {
  async run(event: HealthEvent, step: HealthStep) {
    const { name } = event.body;
 
    const greeting = await step.do("Build greeting", async () => {
      return `Hello, ${name ?? "there"}!`;
    });
 
    return { greeting };
  }
}

Two things are doing work here:

• event.body is whatever you POST when triggering the loop. Type it on HealthEvent and you get autocomplete inside run.
• step.do(name, fn) is a checkpointed unit of work. The platform persists the return value, so if a later step fails and the workflow retries, this one doesn’t re-run. The name is what shows up in the run timeline.

2. Deploy

Click Deploy. The platform stores a new loop_versions row, marks it active, and routes you to the loop’s overview page.

Every deploy is its own version. Older versions stay around under Version history — you can roll back without losing the code.

3. Trigger a run

You have two paths.

From the UI: click Run on the overview page. A dialog lets you paste a JSON body for event.body. Submit it; the run starts immediately and you’ll see it appear under Runs.

From the API: mint a token under Settings → API tokens, then POST to your loop’s trigger URL. The exact URL is on the access dialog — it looks like this:

curl -X POST https://api.ollie.health/loop/$ORG_ID/$LOOP_ID \
  -H "Authorization: Bearer $LOOPS_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{ "name": "Khaya" }'

The response gives you a runId you can use to find the run later.

4. Inspect the run

Open the run from the Runs tab. You’ll see:

• Steps — every step.do that ran, in order, with its status and return value.
• Logs — every console.* your loop wrote, plus any error trace.
• Input / output — what you passed in, and what run() returned.

That’s the full feedback loop: write code, deploy, trigger, read the run. Iterate on the code, deploy again — the next run uses the new version automatically.

Where to go next

• Use connectors to reach outside the loop. The built-in FACIAL_SCAN connector is documented on the overview page.
• Add step.waitForScan when you want the loop to pause until something external happens — the loop suspends without burning compute, then resumes with the results.
• Set up scheduled triggers under Settings to run a loop on a cron.