Data

Bring your own data so your agents work with your information, not just the open web.

Most useful work starts with data the agents don't already have: a spreadsheet of customers, a folder of internal docs, an analytics export, a knowledge base. The Data tab is where you upload that material, organize it, and make it available to your team. Once it's here, any agent on any task can query it.

Three kinds of data

The Data tab is one list with three filters:

• Datasets - tabular data you can query with SQL. Upload CSV, TSV, or Parquet files; Busy Bee parses them, infers the schema, and exposes them to agents via DuckDB. Great for "summarize last quarter's pipeline" or "find every customer who churned in March."
• Documents - long-form text. PDFs, Word docs, Markdown, transcripts. Each document is chunked and indexed so agents can retrieve the relevant passages on demand. This is your knowledge base.
• Connections - live links to external sources (Google Drive, Notion, S3 buckets, etc.) so the team can pull in fresh data without you re-uploading it every week.

📸 Screenshot: the Data tab with the All / Datasets / Documents / Connections filter row visible.

Uploading a dataset

Drag a CSV onto the upload zone (or click + Add) and Busy Bee handles the rest. The file is parsed, column types are detected, the row count and file size are shown, and the dataset is ready to query within seconds. Open a dataset to see its schema, a sample of rows, and the SQL agents will use to reach it. Bigger files take a moment longer; the status badge shows progress.

📸 Screenshot: a dataset detail view with schema, sample rows, and row count.

Knowledge bases

Documents work a little differently. Upload a PDF or a Word file and Busy Bee chunks it into retrievable passages, embeds each chunk, and indexes the result. Agents query that index by meaning, not by keyword, so they can answer questions like "what did we promise about uptime in the MSA?" without reading the entire contract.

Group related documents into a knowledge base and an agent can search across the whole set in one shot.

📸 Screenshot: a knowledge base with several documents listed and chunk counts visible.

Using data in a task

Once data is in, no special syntax is required. Ask an agent to "find the top 10 customers by ARR in the Q3 export" and the agent will pick the right dataset, write the SQL, and return the result. Ask "what's our policy on refunds?" and the agent will hit the knowledge base, pull the relevant passages, and answer. The data layer is invisible to the request, that's the point.

Where to go next

Once your team is producing work, you'll want to see what they've done. That's Analytics.