Large repositories

Performance

Understand the benchmark story, analysis model, and tuning options for large TypeScript codebases.

Benchmarks

Representative local runs against pinned fixtures show Codescythe focusing on files and exports with a smaller runtime profile than broader project-audit tooling.

microsoft/vscode9,398 files
1.11s4.22s
grafana/grafana8,358 files
833ms9.51s
elastic/kibana90,931 files
13.61s43.04s
renovatebot/renovate2,456 files
154ms900ms

Run the same harness locally with pnpm benchmark or target one fixture with pnpm benchmark:kibana.

Analysis Model

Codescythe discovers the project file set, parses files in parallel graph-frontier batches from configured entries, and avoids framework-plugin discovery.

Tuning

Use CODESCYTHE_PARSE_THREADS to tune parse parallelism. RAYON_NUM_THREADS is respected when the Codescythe-specific variable is unset.

CODESCYTHE_PARSE_THREADS=8 npx codescythe --json --config codescythe.jsonc

Repeatable Runs

For stable cleanup automation, keep config checked in, run the same command locally and in CI, and rerun after fix mode because deletions can expose additional unused code.