Skip to main content
A dataset is a curated collection of examples your team wants to keep: events, traces, and custom uploaded rows. Use datasets to hold eval sets, golden examples, labeled training rows, or any slice of traffic you don’t want to hunt for again.

​
How datasets work

Each dataset belongs to one project and holds members of three kinds:
  • Events saved from the Events page
  • Traces saved from a trace view
  • Custom rows uploaded as a file
You can annotate members with labels (Good, Bad, Needs review, Interesting, or your own). Opening a dataset gives you a workspace to filter items, label them, export CSV, and optionally chat with the Triage Agent about that collection. In the All projects view you can browse datasets across active projects, but creating one asks which project it should live in, and uploads / edits stay project-scoped.

​
Using datasets

​
Create a dataset

On Datasets, create a shared collection and give it a name (and optional description). If you’re in All projects, pick the target project first.
Datasets list

​
Save events or traces into a dataset

From an event or trace, use Add to dataset to save it into an existing collection or create a new one. You can also add many matching items at once from a filtered list.
Add event to dataset

​
Upload custom rows

Upload a file into a dataset when the examples don’t already live in Raindrop as events. Uploads require a concrete project (not All projects).

​
Annotate and export

Open a dataset, label items as you review them, remove members you don’t need, and export the collection as CSV when you want it outside Raindrop.

​
Ask Triage about a dataset

With the dataset open, chat with Triage to find more examples like the ones you’ve kept, summarize the collection, or suggest removals.

​
Troubleshooting

I can’t upload or edit in All projects. Switch into the owning project. Org-wide view is for browsing, not writes. Create asks me to pick a project. You’re in All projects. Choose which project the dataset should belong to.
  • Events to find and save examples
  • Signals when a labeled set becomes a tracked behavior
  • Triage Agent to chat about a dataset
  • MCP overview (list_datasets, create_dataset, update_dataset_members, annotate_items)
  • Projects for how datasets are scoped