Data
Bring your own datasets, documents, and connections so your team works with your material, not just the open web.
Your agents are only as good as what they can see. Upload your spreadsheets and files here, and every agent in your workspace can work with your numbers and your documents — not just whatever's on the public web.
Two kinds of data
What you upload is sorted automatically by file type into one of two forms:
- Datasets — tabular files (CSV, TSV, JSON, XLSX, Parquet) become SQL-queryable tables. Agents run queries against them, scanning only the columns and rows a question needs, so it stays fast even on large files. Agents can also write their own results back as new datasets.
- Documents — DOCX, TXT, and Markdown files are split into passages, indexed, and searched by meaning. When an agent needs context, it pulls the most relevant passages instead of the whole file.
Datasets are for questions like "what were Q3 sales by region." Documents are for "what does our refund policy say." Most real work uses both.
What agents do with it
Everything you upload is available to every agent in your workspace — no per-task setup. Agents query datasets to run calculations, spot trends, and build reports, and search documents to ground their answers in your policies, contracts, and reference material.
Connections
Live connections to sources like PostgreSQL, MySQL, and Google Sheets are on the way, so agents can pull current data instead of a one-time upload. For now, upload the files you want agents to work with.
Where it fits
- Datasets & documents — upload files and see how each type is handled
- Connections — external data sources
- Data & analysis — how agents query your datasets
- Knowledge base — how agents search your documents
- Memory — what agents carry forward on their own, separate from the data you upload