Datasets & documents
Upload your spreadsheets and files so your team works with your data, not just the open web.
Upload your own data and files so your team reasons over your material — your customer list, your revenue export, your contracts — instead of guessing from the open web. Every file you add is available to all agents in your workspace.
Two kinds of data
The upload sorts each file automatically by its extension:
- Datasets — tabular files (CSV, TSV, JSON, JSONL, XLSX, XLS, Parquet). These become queryable tables. Give each one a short name (letters, numbers, underscores) that your team uses to reference it; the display name is optional and just for people.
- Documents — PDF, TXT, Markdown, DOCX. These get read, split into passages, and indexed so agents can search across them.
Datasets accept files up to 100 MB; only the tabular types above are allowed. Documents cap at 10 MB, and executable files are rejected.
How your team uses it
Datasets are queried with SQL. An agent asked to "find the top 10 customers by spend last quarter" writes a query against your uploaded table and gets exact numbers back — only the columns and rows it needs are scanned, so it stays fast even on large files. See data & analysis.
Documents are searched by meaning. Each document is chunked and embedded, so an agent retrieves the most relevant passages for a question rather than needing an exact keyword match. This powers your knowledge base.
Datasets your agents create
When an agent produces structured output — a cleaned export, a computed table — it's saved back here as a dataset, marked as agent output. That means one agent's results become queryable input for the next.
Watching it come in
Newly uploaded data shows a Processing status while it's parsed and indexed, then flips to Ready (or Failed, with the reason). You can preview a dataset's rows and columns, or read an auto-generated summary of a document, from its detail view.
Live database connections (PostgreSQL, MySQL, Google Sheets) are coming soon.
Where it fits
- Data & analysis — how agents query your datasets
- Knowledge base — how agents search your documents
- Connections — connecting live data sources