Your first workflow: from a page URL to a useful summary
Use the page-summary template to learn inputs, connected nodes, and output review without building a graph from scratch.
By Hot wasl
Published
A useful first workflow has one input and an output you can check yourself. The “Summarise any page” template does exactly that: accept a URL, read the page, summarise its text, and return a result named summary. Start with a public article you have already read so you can judge the result rather than just admire a successful status.
Install a small, understandable graph
Open the template linked below and choose “Use template”. If you are not signed in, create an account first. Installing creates a workflow in your workspace and opens the builder. The starting graph has four nodes: a manual trigger, Read web page, Summarise, and a result output. Keep that simple shape for your first test.
- Check that the manual trigger collects an input named url.
- Inspect Read web page: its URL comes from {{$trigger.url}}. This template limits extracted text to 12,000 characters.
- Inspect Summarise: its source is {{read.text}}, its style is bullets, and its language setting is automatic.
- Keep the output name summary so the final result is easy to identify.
Run one page, then read the evidence
In the builder’s run panel, enter the complete public URL in the url field and start the run. You need a positive workspace credit balance to start a new run. Web page reads consume platform credits; an attached personal model key only removes platform-credit charges for the model call, and the provider bills separately.
Open the execution trace and expand the reading node before judging the summary. Look at its title and text. A page that relies on browser scripts, requires a login, or blocks automated requests may not supply the article you expected. A successful fetch is not proof that the right content was extracted.
Improve the result one change at a time
Compare the summary with the source: are the main claims present, are important qualifications retained, and has anything unsupported appeared? If the text was clipped, the summary cannot cover the missing portion. Adjust the reading limit or choose a shorter source before changing the summarisation settings.
Watch for a simulated-model notice in the trace. When no model key is configured, the app can return a simulated response; use it to inspect the workflow, not as evidence of a live model’s quality. Once the basic path works, change one setting and rerun the same URL for a meaningful comparison.
- The extracted text contains the article, not a sign-in screen.
- The summary preserves the source’s meaning and language.
- The final outputs contain summary, and the trace explains the credit usage.