Skip to content

Research overview

Research is the umbrella for systematic work in Chatty the Lab. It brings your datasets, experiments, human-labeling jobs, and research notes together so a line of inquiry stays organized from raw data to written-up findings.

Use Research when you want to go beyond one-off chats: run a parameter sweep over an agent graph, annotate a dataset with a team, compare results across trials, and keep a shared logbook of what you learned.

The hub (/research) lists two things:

  • Investigations — research containers, each showing its counts of experiments, annotation jobs, and datasets.
  • Datasets — your global, shareable data tables (also reachable from any investigation).

An investigation is a soft grouping around a question or project. It is intentionally lightweight:

  • It owns its experiments and labeling jobs (1:N) — they belong to exactly one investigation.
  • It shares datasets (M:N) — the same dataset can be used by many investigations and across workspaces.

You can create, rename, and delete investigations at any time. Deleting one does not delete the datasets it merely shared.

Open an investigation (/research/[id]) to find six tabs:

TabWhat it holds
OverviewSummary of the investigation, its counts, and quick links
DatasetsDatasets attached to this investigation (shared, M:N)
ExperimentsParameter sweeps over an agent graph, producing trials
AnnotationsHuman-labeling jobs over a dataset
ResultsAggregated outcomes and plots across experiments and trials
LogbookCollaborative markdown notes and findings

A typical flow: attach a dataset, build an experiment that runs an agent graph over that dataset across a parameter space, inspect the resulting trials and results, optionally spin up a labeling job to have humans annotate records, and write everything up in the logbook — linking back to the specific experiments and trials you reference.