Summarize a collection
Summarizer produces one summary per document, configurable length and tone, downloadable and savable.
Summarizer runs the same summarization task across every source in a dataset. One summary per document, the way you specify, all downloadable in a single run.
What you get
- One summary per source in the dataset.
- Each summary uses the length and tone you picked.
- Downloadable individually (as PDF or markdown) or all at once as a ZIP.
- Optionally saved into your workspace as notes (one note per source) for further editing in Co-work.
Run it
-
1
Build the dataset: pick a label, multi-select on the documents page, or use Smart Search to find the right set.
-
2
Open the workflow runner and pick Summarizer.
-
3
Set the inputs (see below).
-
4
Click Run. The workflow dispatches a sub-worker per source and persists the results to a durable record. You can leave the page; the run continues in the background.
Inputs
- Length: brief (a few sentences), medium (a paragraph or two), detailed (multi-section). Brief is faster and cheaper; detailed catches more nuance.
- Tone: neutral, formal, conversational, technical. Pick what fits the audience.
- Format: narrative prose or bulleted. Bulleted is easier to skim; narrative reads better.
- Custom instructions: optional free-form guidance ("focus on financial risks", "ignore boilerplate sections", "include section page numbers").
The custom-instructions field is the most under-used and most useful input. A short steer ("emphasize renewal terms and termination clauses") changes the summaries noticeably.
When summaries land
Each per-source summary lands as it completes. You can open partial results while later sources are still running.
If the dataset is large, the workflow respects your plan's prompt fair-use limits. Workflow runs are credit-metered; see How credits work.
After the run
- Save as notes: turns the summaries into editable notes in your workspace, useful when you want to refine some of them in Co-work.
- Download: pull a ZIP of all summaries, or grab individual PDFs.
- Open the run record: the workflow lives in your workspace's runs list; re-open it any time.
Common patterns
- Onboarding a stack of contracts: Summarizer with
length: medium,format: bulleted, and an instruction like "focus on parties, term, renewal, termination, and governing law." - Triaging a research dump: Summarizer with
length: briefto find the few papers worth reading in full. - Briefing for a meeting: Summarizer with
format: bulletedand "limit to facts the attendees won't already know."
What's next
- Extract structured data: if you want the same fields across documents, not summaries.
- Capture context with notes: refine saved summaries in Co-work.
- How credits work.