Import into BigQuery — CSV, Excel & PDF

• Overview
• From
• Into

Importing for your users — or for yourself?

Building a product or data platform on BigQuery? Customer uploads or analyst spreadsheets — this page is your integration guide: typed, validated rows up front, and your handler owns the load. Read on.

Just trying to load your own file into BigQuery? You have the file and the dataset; the file is the problem — renamed headers, merged columns, a PDF where a spreadsheet should be. See the wizard handle genuinely messy files in the playground (free, no signup). And a run-it-yourself importer is in the works: your own Google Cloud credentials, runs on your machine, every row staged for review — and revertible after it lands. Join the waitlist.

Goal: your users bring the file they have — CSV, Excel, even a PDF or a scan — and clean records land in BigQuery without leaving your product, and without you writing a mapping/validation UI. The pattern is the same every time: embed → validate → deliver. Mildport handles the first two; your webhook handler owns the write into BigQuery.

1. Embed the importer

<mildport-import
  api-base-url="https://imports.your-infra.example"
  license-key="SIGNED_TENANT_KEY"
  catalog-key="events"
></mildport-import>

2. Publish a catalog that mirrors BigQuery

The schema lives in a target catalog, not in host code. Create it in the catalog editor (/admin/catalogs on your engine, or embed <mildport-catalog-editor>), match the destination table, and publish. The widget loads that published catalog by catalog-key.

Align keys with column names (user_email, event_name, amount, occurred_at). Mark REQUIRED columns required; the review step then refuses rows BigQuery would reject.

Users then upload anything — renamed headers, six languages, XLSX/ODS, even PDF tables — and Mildport’s deterministic matcher plus the review step produce typed, validated rows.

Need the schema computed per user at runtime, or versioned with your app? That’s the schema-as-code path — pass targets from host code instead of catalog-key.

3. Deliver from your webhook

Clean rows arrive on your signed apply webhook; your handler writes to BigQuery:

Batch load job (GCS → LOAD) for bulk — atomic and free — or the Storage Write API for streaming

BigQuery-specific gotchas

• prefer load jobs over streaming inserts for imports: atomic, free, and no best-effort dedup window to reason about.
• REQUIRED schema fields fail the whole load — mirror them as required catalog fields so review guarantees loadability.
• partition by ingestion time or a date column up front; retrofitting partitioning means rewriting the table.

Why not hand-rolled import?

bq load assumes an engineer and a clean file. When analysts or customers bring spreadsheets, an embedded importer types and validates every row first — so a load job never burns an hour on row 40,213’s date format.

---

Try the flow first: the playground (no signup), then the Quickstart and target catalogs. Prices are published — hosted and flat-annual self-host — at mildport.com/#pricing.