Typesafe CQL.
Written in Rust.

Seven small tools that move mistakes and slowdowns in Cassandra and ScyllaDB queries from production to compile time and CI: typed, checked against current Cassandra, and measured for performance.

Get started Benchmarks Examples

$ cargo install cqlc

The family

Seven tools, one foundation

All of them build on the ScyllaDB Rust driver and cqlc-core, the lexer, parser, schema resolver and type checker that checks every statement.

Schema-first

cqlc

Write .cql schema and named queries. cqlc type-checks them offline against Cassandra 4.0 to 5.0 CQL (vectors, SAI, ANN search) and generates typed async Rust.

3.8×faster parsing than sqlparser on identical statements

Code-first

cqlx

Write a Rust struct. #[derive(CqlEntity)] generates the table, inserts and a type-state query builder that cannot forget a partition key.

95–97 %of the raw driver on YCSB; 1.0 with partial updates, LWT, batches, streaming

Performance

cql-ai

A 0.6M-parameter Transformer that predicts performance bottlenecks from the statement and your data shape, trained on measured latencies of public workloads.

0.86macro-F1 on unseen data shapes; keyword rules reach 0.56

Design

cqlz-gym

Give it entities and a query workload; it designs the Cassandra tables. An exact ILP and a small learned policy search, cqlc-core verifies every candidate.

+8 %+8 %SMALL

Lint + format

cqlint

A ruff for CQL: formats your files and flags ALLOW FILTERING, tombstone writes, unbounded scans and more. Type-aware with your schema.

0.24 sto lint 3,600 files (23 MB)

Migrations

cql-migrate

Ordered, checksummed, locked schema migrations with offline chain checks, schema diff and live drift detection.

2 engineslifecycle tested on ScyllaDB and Cassandra 5.0

Profiling

cqlprof

EXPLAIN for CQL: runs a SELECT with tracing and tells you about tombstones, sstables, replicas and tail latency.

PROF001found 2,990 tombstones behind 10 live rows

Measured, not claimed

Parsing CQL, on identical statements

Time to parse the same 22 statements (lower is better)

cqlc parser

33 µs

sqlparser

126 µs

cqlc's complete analysis, which also type-checks every query against the schema, takes 103 µs for all 36 statements. sqlparser parses SQL, not CQL, so it only accepts 22 of them. All benchmarks and caveats →

How they fit together

One parser under everything

.cql filesschema + queries

→

cqlc-corelexer, parser, resolver, type checker

→

cqlctyped Rust code

Rust structs#[derive(CqlEntity)]

→

cqlxtype-state builders, CREATE TABLE

→

cqlx schemaCQL checked by cqlc-core

.cql filesqueries in your repo

→

cqlc-core lexerfixed integer ids

→

cql-aiTransformer, CI warnings

entities + workloadTOML

→

cqlz-gymILP and learned policy, verified by cqlc-core

→

schema.cqlfeeds cqlc and cqlx

Typesafe CQL.Written in Rust.