Schedule a daily report with an LLM summary

Recurring reports are where metered platforms quietly get expensive: the schedule itself consumes quota all month. In Pipory the waiting is free, so a daily digest is one of the most natural workflows to build. This one fetches data, summarizes it with a model, and posts the digest to Slack every morning.

If this is your first workflow, start with webhook to Slack - it introduces the canvas and templates this guide builds on.

Add a Schedule trigger

Create a workflow with a Schedule node as its trigger. Give it a cron expression and a timezone:

0 8 * * *

That fires at 08:00 every day in the timezone you pick. The node has an enable toggle - leave it off while you build, and use a manual run to test instead. Under the hood one engine-side cron ticks every minute and fans out to whichever schedules are due, so editing the expression takes effect without any re-registration.

Fetch the data

Add an HTTP Request node pointing at whatever you want to report on - your product's metrics endpoint, an issue tracker's API, anything that returns JSON. Give the node a variable name like report so downstream nodes can read the response from the context.

Summarize with a model

Add an Anthropic (or OpenAI or Gemini) node using your own API-key credential. Template the fetched data straight into the prompt:

Summarize the following metrics for a busy founder in five bullet points.
Highlight anything unusual: {{json report.httpResponse}}

The model's reply lands in the context under the node's variable name, ready for delivery.

Deliver it

Add a Slack node (or the Email node with a Resend credential) and template the summary into the message. Run the workflow manually once to check the output reads well, adjust the prompt, then enable the schedule.

Notes from real use

• The {{json ...}} helper serializes an object into readable JSON inside a template - useful when passing structured data to a model.
• Every morning's run is recorded with per-node input, output, and duration, so a bad digest is a two-minute diagnosis: open the execution and read what the model actually received.
• Disabling the workflow stops future runs immediately; the schedule simply skips it.