cube Sign in
Connectors

Point cube at anything.

One connection URI is the entire setup. cube detects the source, opens it read-only, and derives the rest itself. No schema registration, no modelling files, no agent to install.

6
sources live and verified
read-only
on every source
your VPC
runs inside your perimeter

Every connector is read-only by construction. A URI that demands write scope is rejected at the door. In enterprise deployments the whole engine runs inside your VPC: your data never leaves your walls; what compounds is how cube reads lakes, not what is in yours.

  • Postgres

    LIVE · VERIFIED. Live and verified against a real server in CI

    Verified against a real server in CI.

    postgres:// postgresql://
    • A read-only role. SELECT and USAGE only; cube asserts it cannot write before it reads anything.
    • A connection URI: postgres://cube_ro@host:5432/db.
    • Prefer a read replica. Profiling is sampled and bounded, and a replica keeps it off your primary entirely.
    read-only · in your perimeter · PII classified and excluded
    Connect your lake →

  • MySQL / MariaDB

    LIVE · VERIFIED. Live and verified against a real server in CI

    Verified against a real server in CI.

    mysql:// mariadb://
    • A read-only user with SELECT on the target schema.
    • A connection URI: mysql://cube_ro@host:3306/db.
    • Prefer a replica endpoint for the same reason as Postgres.
    read-only · in your perimeter · PII classified and excluded
    Connect your lake →

  • Object store: S3, GCS, Azure, R2

    LIVE · VERIFIED. Live and verified against a real server in CI

    Verified against a real server in CI.

    s3:// gs:// gcs:// az:// azure:// r2://
    • Read-only credentials scoped to a prefix: GetObject and ListBucket, nothing else.
    • A URI to the data: s3://bucket/lake/ · parquet, csv, json, arrow.
    • Partitioned layouts like dt=2024-01/ are understood as one relation.
    read-only · in your perimeter · PII classified and excluded
    Connect your lake →

  • Files: a directory or glob

    LIVE · VERIFIED. Live and verified against a real server in CI

    Verified end to end on every test run.

    file:// · parquet csv tsv json ndjson arrow
    • Read access to a path. Parquet, CSV, JSON, Arrow, mixed freely.
    • Sharded files like sales/2024-01.parquet, -02, -03 collapse into one relation.
    read-only · in your perimeter · PII classified and excluded
    Connect your lake →

  • DuckDB

    LIVE · VERIFIED. Live and verified against a real server in CI

    Verified end to end on every test run.

    duckdb:// · duckdb ddb
    • Read access to a .duckdb file. cube opens it read-only.
    read-only · in your perimeter · PII classified and excluded
    Connect your lake →

  • SQLite

    LIVE · VERIFIED. Live and verified against a real server in CI

    Verified end to end on every test run.

    sqlite:// · sqlite sqlite3 db
    • Read access to a .sqlite file. Opened read-only, same contract as every other source.
    read-only · in your perimeter · PII classified and excluded
    Connect your lake →

  • Snowflake

    ON THE ROADMAP. On the roadmap, contract complete, not yet connectable

    Contract complete. Not yet connectable.

    snowflake://
    • Will read INFORMATION_SCHEMA through a read-only role into the same profile shape. The contract is written; the wiring is not. We do not connect what we have not verified.
    Tell us this one matters →

  • BigQuery

    ON THE ROADMAP. On the roadmap, contract complete, not yet connectable

    Contract complete. Not yet connectable.

    bigquery:// bq://
    • Will read table metadata with a dataViewer-grade role. Same contract, same posture, not yet live.
    Tell us this one matters →
connect

One read-only URI. cube detects the source and asserts it cannot write.

profile

Tables, keys, cardinality, and the join graph, recovered from the data itself.

infer

Entities, measures, dimensions, grains. The semantic model, derived, reviewable.

precompute

The bounded ask-space built once, the honesty rails folded into every cell.

serve

Microsecond answers with the base attached. Below the floor: refused, never guessed.

The full pipeline, the honesty model, and the trust states live in the docs. Read the docs →

Connect your lake, free.

Sign in with a work email, and the first mile is free: profile, inferred model, generated ask-space, first answers with the honesty rails on. First-mile runs are scheduled in order.

Sign in

cube is built for lakes with hundreds of tables and billions of rows. A laptop CSV is not the customer.