Skip to main content
A dataset is a table Vistazo extracted from a file somebody posted: a CSV, a spreadsheet, a table inside a PDF, a screenshot of a report. Once extracted it behaves like any other data source. A widget can be wired to it, charted, and refreshed. This is the answer to “can I just paste a CSV here?”.

Getting data in

Attach a file to a chat or a discussion (Attachments). Extraction happens on upload, and the resulting tables land in your workspace’s datasets. There is nothing to configure. Ask for a chart of the file you just posted and the agent finds it.

Datasets are not a connection

You will not find a “Datasets” row under Settings → Connections, and that is deliberate. Being asked to connect to your own files would be odd. The tools that read them are always available:
  • on Vistazo’s core MCP surface, so your own agent can read your uploaded tables with no second authorization step;
  • to widget pipelines, so a chart can be wired to a dataset and refresh against it.
See the dataset tool reference.

Who can read one

A dataset is readable by anyone who can read the thread the file was posted in, and by anyone who can view a dashboard it has been wired into. That second part is what makes uploaded data usable on a shared dashboard, and it is worth reading twice before you chart something sensitive. Binding a table from a private chat into a shared dashboard gives that dashboard’s viewers the table, and the file behind it. Datasets are workspace-scoped, and they are the only Vistazo-hosted source that holds your data rather than public data.

Storage and lifecycle

If your storage is full when a file is uploaded, the upload still succeeds and its text still lands, but the tables that do not fit are not saved as datasets. Free up space and upload again to get them. Deleting the attachment stops reads and credits the bytes back to your quota. A dataset that no dashboard uses goes with the file. One that a widget is still wired to leaves a marker behind instead, so the chart can say Source file deleted and offer to be pointed at replacement data, rather than failing with nothing to explain it.