One binary, two modes
Embedded single-process engine and distributed coordinator + stateless workers — the same SQL surface, the same binary.
Sovereign Query Engine Iceberg-first SQL
Iceberg-first SQL
that scales.
Run SQE embedded as one binary on your laptop, or distributed across a cluster of stateless workers behind Arrow Flight SQL or Trino HTTP. Same SQL surface, same Iceberg semantics, same identity model. Every query runs as the authenticated user — no service account, no shared root.
Build from source: cargo install --path crates/sqe-cli — Apache-2.0, Rust, no JVM.
# Point SQE at AWS S3 Tables and run it as a SQL server. $ sqe-cli --embedded SQE 0.15.0 · DataFusion 53.1 · ~/.sqe/warehouse/ # Iceberg time travel + per-query identity. sqe> SELECT customer_id, sum(amount) ...> FROM s3tables.sales.orders ...> FOR TIMESTAMP AS OF '2026-04-01'; +-------------+-------------+ | customer_id | sum(amount) | +-------------+-------------+ | 100482 | 18204.55 | +-------------+-------------+ 12 rows · 47ms
Why SQE
The lakehouse engine you actually own.
88.4%
Top-five on the public Iceberg matrix
167/189 cells. The only non-Spark engine in the top five — scored against the public rubric, not a vendor spec sheet.
See the matrix →6/7
Benchmark suites won vs Trino 465
222 of 222 queries pass at SF1. Up to 18.5× faster on TPC-E. Reproducible with one script.
See the receipts →0
Service accounts. Zero shared roots.
Every query runs as the authenticated user. The bearer token flows through to the catalog and storage. Policy is enforced before the optimizer runs.
vs Trino →1
Binary, laptop to cluster
A DuckDB-class embedded engine and a distributed coordinator share one SQL surface.
Cross-catalog joins, runtime ATTACH, persistent local warehouse.
What it is
Everything in one engine.
Native Iceberg V2 + V3
Read and write, not read-only. Position & equality deletes, merge-on-read, copy-on-write, schema evolution.
Per-query identity
OIDC password grant; the user’s bearer token is passed through to the catalog and S3 on every query.
Policy as plan rewriting
Row filters and column masks via OPA or Cedar, enforced at the LogicalPlan layer before the optimizer runs.
7 catalog backends
Polaris, Nessie, Unity Catalog OSS, AWS Glue, S3 Tables, Hive Metastore, JDBC — plus Hadoop storage. Multi-cloud.
Two wire protocols
Arrow Flight SQL and Trino HTTP from the same coordinator. Connect existing Trino clients unchanged.
Rust, no JVM
Sub-second cold start. A C++/DuckDB-class footprint with a distributed story the JVM engines charge for.
Column-level lineage
Native OpenLineage 2-0-2 emitter with column-level lineage on writes. File and HTTP sinks, disk-spool fallback.
How to run it
Two commands to a SQL prompt.
Embedded — one binary, no cluster
cargo install --path crates/sqe-cli
sqe-cli --embedded # persistent warehouse at ~/.sqe/warehouse/
sqe> SELECT * FROM '/data/sales.parquet' LIMIT 5;
sqe> SELECT * FROM read_csv('s3://bucket/orders.tsv.gz');
sqe> ATTACH 'http://catalog.example.com/api/catalog' AS partner_cat
(TYPE iceberg_rest, WAREHOUSE 'analytics');
sqe> SELECT * FROM partner_cat.sales.orders LIMIT 10;Cluster — Polaris + S3 + SQE locally
docker compose -f docker-compose.test.yml up -d
./scripts/bootstrap-test.sh
cargo run --release --bin sqe-coordinator -- tests/sqe-test.toml
# Connect with the CLI over Arrow Flight SQL
cargo run --bin sqe-cli -- --host localhost --port 50051 \
--username root --protocol flight
sqe> SHOW CATALOGS;
sqe> SELECT * FROM test_warehouse.default.my_table LIMIT 10;Full quickstart covering all seven catalog backends → docs.getsqe.com
Receipts
How SQE differs.
| SQE | Trino | DuckDB | |
|---|---|---|---|
| Embedded mode (one binary) | yes | no | yes |
| Distributed mode | yes | yes | no |
| Iceberg V2 + V3 read + write | native | V2 + partial V3 | read-only |
| Per-query OIDC passthrough | yes | service acct only | n/a |
| OPA / Cedar at LogicalPlan | yes | no | no |
| Multi-catalog in one engine | 7 backends | one at a time | per-extension |
| Wire protocols | Flight SQL + Trino | Trino HTTP | extension |
| Runtime | Rust, no JVM | JVM | C++ |
| Cold start | sub-second | tens of seconds | sub-second |
| OpenLineage emitter | native, column-level | plugin | no |
Performance — SF1, vs Trino 465
| Suite | SQE | Trino | Speedup | Pass |
|---|---|---|---|---|
| TPC-H (22) | 16.8s | 26.7s | 1.6× | 22/22 |
| SSB (13) | 8.3s | 5.8s | 0.70× slower | 13/13 |
| TPC-DS (99) | 13.4s | 45.6s | 3.4× | 93/99 |
| TPC-C (8 read) | 0.41s | 2.65s | 6.5× | 8/8 |
| TPC-E (11) | 9.3s | 172.0s | 18.5× | 11/11 |
| TPC-BB (10) | 28.0s | 255.7s | 9.1× | 10/10 |
| ClickBench (43) | 1.3s | 4.46s | 3.4× | 43/43 |
SQE wins six of seven suites. 222/222 queries pass. SSB is the one suite SQE trails —
lineorder’s uniform FK distribution defeats row-group pruning. Run your own with
BENCH_SCALE=1 ./scripts/benchmark-test.sh --compare-trino.
Active development
Recently shipped
- Jun 2026 Read-only ops web UI, in the binary
- May 2026 DuckDB-class embedded mode — file TVFs,
hf://, Delta reader - May 2026 Column-level OpenLineage emitter