Connect data sources

Insights Engine answers questions by reading sources. A source is any file, table or feed of data you've connected to a project. The most common starting point is a file upload, but several other source types are supported too.

Open the Sources page

From the sidebar, click Sources. You'll see four tabs across the top:

• All — every source in the project, with status, row count, and freshness.
• Upload — drag-and-drop CSV / Excel files.
• Apps — third-party app connectors.
• BigQuery — direct BigQuery table connections (advanced).

Uploading a file

On the Upload tab, drag a CSV or XLSX file onto the drop zone — or click to pick one. You can drop several files at once.

Each file goes through these stages, shown live on a status badge:

1. Queued — waiting for a slot.
2. Uploading — the file is being transferred securely.
3. Parsing — the platform reads the rows and columns.
4. Analyzing — every column is profiled (data type, sample values, hints about what it represents) and matched to known business concepts.
5. Indexing — the source is added to the project's question-answering index.
6. Ready — you can ask questions about it.

If something looks off — for example, a date column the platform can't parse — you'll see a gotcha chip on the source with a short explanation and a suggested fix.

Connecting BigQuery

If your data already lives in BigQuery, you can connect a dataset directly without exporting files. On the BigQuery tab, paste your project + dataset reference and the platform will inspect available tables for you to import.

When you connect a BigQuery table, the platform reads its schema and a representative sample — it does not download every row. Queries run live against your warehouse.

App connectors

The Apps tab lists third-party services you can connect — e.g. Google Drive for documents, file shares for ongoing uploads. New connectors are added regularly; if you don't see one you need, ask your admin.

Trust, scope and versions

Every source has three properties worth knowing about:

• Trust level — trusted (use freely in answers) or sandbox (visible to you but not auto-cited until you promote it).
• Scope — owned (this project owns it) or subscribed (shared from another project in the same workspace).
• Versions — every re-upload is captured as a new version. The platform shows the schema diff between versions so you can see what changed.

This matters because every answer is cited back to a specific source version — so an answer you got last month is still defensible against the snapshot of data you had then.

What about live data?

For files: re-upload to refresh. For BigQuery: queries run live. For app connectors: each connector sets its own refresh cadence, shown on the source.

Tips

• Name files descriptively. 2026_q1_deals_export.csv beats data.csv — the platform uses filename hints to auto-map columns.
• Don't pre-aggregate if you can avoid it. Raw rows give the platform more to work with than summary tables.
• Watch for gotchas. They surface quickly and usually take seconds to fix.

Next: Ask your first question.