ql

ql lets developers query codebases with SQL.

It parses source files with Tree-sitter, maps syntax into language-agnostic tables, loads rows into DuckDB, and runs SQL against those tables. Goal: deterministic code search and analysis without AI, embeddings, or fuzzy guesses.

From a checkout:

cargo build --release
./target/release/ql --langs

Or install the binary locally:

cargo install --path crates/ql-cli
ql --langs

ql "SELECT name, file, line
    FROM functions
    WHERE complexity > 10
      AND has_test = false" ./src

Language adapters populate the shared schema:

• functions
• calls
• imports
• structs
• variables
• comments
• fn_fingerprints — structural metrics per function
• fn_callsets — callee sets per function
• similarities — pairwise similarity scores

Schema source lives in schema/tables.json.

All examples assume you run them from a repository root or pass an explicit path as the final argument.

1. Find non-test functions with higher complexity.

ql "SELECT name, file, line, complexity FROM functions WHERE has_test = false AND complexity > 10 ORDER BY complexity DESC" .

2. List all external calls.

ql "SELECT caller, callee, file, line FROM calls WHERE is_external = true ORDER BY file, line" .

3. Inspect imports from the standard library.

ql "SELECT module, alias, file, line FROM imports WHERE is_std = true ORDER BY file, line" .

4. Find public structs and their implemented interfaces.

ql "SELECT name, visibility, implements, file, line FROM structs WHERE visibility = 'public' ORDER BY file, line" .

5. List mutable variables.

ql "SELECT name, type_hint, scope, file, line FROM variables WHERE is_mutated = true ORDER BY file, line" .

6. Show doc comments attached to code.

ql "SELECT text, attached_to, file, line FROM comments WHERE is_doc = true AND attached_to != '' ORDER BY file, line" .

7. Find functions that mention tests.

ql "SELECT name, file, line FROM functions WHERE has_test = true ORDER BY file, line" .

8. Show functions that return results.

ql "SELECT name, return_type, file, line FROM functions WHERE return_type LIKE '%Result%' ORDER BY file, line" .

9. Render the same data as JSON.

ql --format json "SELECT name, file, line FROM structs ORDER BY file, line LIMIT 20" .

10. Render a compact CSV export.

ql --format csv "SELECT name, file, line FROM functions ORDER BY file, line LIMIT 50" .

11. Find structurally similar functions (code clones).

ql "SELECT name_a, file_a, name_b, file_b, combined_score FROM similarities ORDER BY combined_score DESC LIMIT 20" .

12. Find functions similar to a specific one.

ql "SELECT name_b, file_b, combined_score FROM similarities WHERE name_a = 'parse_config' ORDER BY combined_score DESC" .

13. Find functions with similar call patterns.

ql "SELECT s.name_a, s.name_b, s.behavioral_score FROM similarities s WHERE s.behavioral_score > 0.8 ORDER BY s.behavioral_score DESC" .

14. Inspect structural fingerprint of a function.

ql "SELECT name, file, complexity, nesting_depth, branch_count, loop_count, call_count FROM fn_fingerprints ORDER BY complexity DESC LIMIT 20" .

ql/
├── crates/
│   ├── ql-ast/         AST bridge, Tree-sitter walk, schema mapper
│   ├── ql-adapters/    Tree-sitter adapter implementations per language
│   ├── ql-core/        Query engine, SQL parser, planner, DuckDB execution
│   └── ql-cli/         CLI binary — arg parsing, output formatting, watch mode
└── extension/          VS Code extension (TypeScript)

Single binary. No subprocess protocol. Rust CLI only.

• Shared row types and TableBatch
• Language adapter trait and Tree-sitter walk path
• Go, Rust, TypeScript, and Python adapter support
• DuckDB-backed in-memory schema with multi-table ingest
• Hand-written SQL parser (SELECT, FROM, JOIN, WHERE, ORDER BY, LIMIT, operators)
• Planner maps SQL AST to DuckDB execution
• CLI with table/JSON/CSV output and watch mode
• VS Code extension sidebar for running queries and opening rows
• Structural and behavioral code similarity via function fingerprints and call patterns

cargo test
cargo build --bin ql

v1 targets Linux and macOS. No AI, no remote repositories, no type-resolution-heavy semantic analysis, no plugin system.

• add *.ext search in query
• add line:col in table in cmd
• better UI
• cross-platform