Enabled with --services bigquery

A community BigQuery emulator, clearly labelled.

Datasets, tables, queries, DDL, scripting, loads, the Storage APIs and BigQuery ML in every Resource Manager project, on CloudBurrow's patched build of the community emulator.

Backed by: goccy/bigquery-emulator 0.8.1, built by CloudBurrow with patches and a validating front Community emulator MIT

Get started
Local scope

What it covers

Start the community emulator with --services bigquery. CloudBurrow builds the emulator from pinned modules with patches to its SQL engine, embeds it in the CLI and puts a validating front before it, which holds IDs, schemas, rows and statements to BigQuery’s rules and answers 501 for what it cannot run. Every project Resource Manager has is served. Remote models, AI functions and image functions run on the instance’s own local AI, never on Google. The compatibility notes distinguish tested SQL shapes, Storage API operations and REST methods. Follow the named tests when deciding whether a query belongs in your local suite. The emulator keeps nothing across a restart.

  • Datasets and tables, streaming inserts (tabledata.insertAll) and SELECT queries
  • Every ACTIVE Resource Manager project is a BigQuery project: datasets, tables and jobs per project, and queries, DML, loads, copies, clones and snapshots across projects (#1764, #1767, #1781)
  • A name without a project resolves in the job's project: tables, views, functions, procedures and models (#1784)
  • Per-project IAM for a caller with an access token this instance minted: jobs, datasets, tables, routines, models and authorized views, with inherited roles and IAM Conditions (#1767, #1774)
  • DML, DDL, views, materialized views, scripts, procedures and multi-statement transactions
  • Partitioned and clustered tables, time travel, change history and INFORMATION_SCHEMA views
  • Loads of CSV, JSON, Avro, Parquet and ORC from uploads or Cloud Storage, and extracts to Cloud Storage
  • The Storage Read API (Arrow and Avro) and Storage Write API streams with protobuf or Arrow rows
  • BigQuery ML: trained models from linear regression to ARIMA_PLUS, and imported XGBoost, TensorFlow Lite and TensorFlow SavedModel models (#1711, #1775)
  • Remote models over the instance's local Vertex AI endpoint, never Google: ML.GENERATE_TEXT and ML.GENERATE_EMBEDDING, with max_output_tokens, temperature, top_p and stop_sequences applied (#1760, #1797)
  • AI functions on the local endpoint: AI.GENERATE with output_schema, AI.GENERATE_BOOL, _INT, _DOUBLE and AI.GENERATE_TABLE through constrained decoding, and AI.GENERATE_EMBEDDING, AI.EMBED, AI.SIMILARITY, AI.CLASSIFY, AI.IF and AI.SCORE (#1771, #1772)
  • Object tables over the instance's Cloud Storage, with metadata caching and the connection's permissions checked (#1783)
  • Image functions over object tables: ML.DECODE_IMAGE, ML.RESIZE_IMAGE, ML.CONVERT_COLOR_SPACE, ML.CONVERT_IMAGE_TYPE, and ML.PREDICT of an imported TensorFlow Lite or TensorFlow image model (#1798)
  • Vision text detection: a CLOUD_AI_VISION_V1 remote model and ML.ANNOTATE_IMAGE with TEXT_DETECTION and DOCUMENT_TEXT_DETECTION, read locally with Tesseract (#1773)
  • The BigQuery Connection API, with connection IAM; remote models and object tables need bigquery.connections.use on their connection (#1755, #1765)
  • Table, routine and row access policy IAM, stored, and generated BigQuery REST API coverage with named tests

In the console

BigQuery datasets in the local console
Captured 2026-09-28 from main @5da2bd0. Product icons hidden.
Release and main

What's verified

Follow the operation matrix to see what the backend serves and the tests linked to Verified calls.

BigQuery (REST API v2)

55 of 55 RPCs Verified in v0.3.0

BigQuery (REST API v2) in v0.3.0: 55 Verified of 55
Inspect the BigQuery (REST API v2) operation matrix →
Local boundaries

Not supported

  • 01 / Local boundary

    BigQuery data is in memory and is gone after a restart, whatever --mode says. A project Resource Manager does not have is 404; create or seed it first.

  • 02 / Local boundary

    Imported ONNX models need the opt-in ONNX build. Vision features other than text detection, the other Cloud AI service types, and safety_settings answer 501 with the reason.

  • 03 / Local boundary

    Stored policies are not checked for requests without a minted token: the ADC fixture's token and unauthenticated calls are served as before.

Evidence source

Source: docs/compatibility.md at main 0ea2594.

Frequently asked

Questions

Is this Google's BigQuery emulator?

No. It is the community goccy/bigquery-emulator. CloudBurrow embeds a patched build behind a validating front and documents the specific verified operations.
Link to this answer

Can I use more than one project?

Yes, since v0.2.0. Every ACTIVE project in Resource Manager is a BigQuery project, and a job in one project can read and write another's tables. v0.1.0 served one project.
Link to this answer

Do remote models and AI functions call Google?

No. They go to the instance's own Vertex AI endpoint: generation runs on the local LiteRT-LM runtime, embeddings on the local embedding model, and Vision text detection on Tesseract.
Link to this answer
Develop locally

Build against Google Cloud APIs, locally

Free and open source under Apache-2.0. No account, no sign-up, no Google Cloud bill.

Get started