Motivation
Graph build and entity extraction latency are common concerns for new users. A small benchmark profile would make the speed-quality tradeoff easier to understand and would give maintainers a stable baseline for future optimization.
Scope
- Add a tiny public fixture corpus for benchmark runs.
- Measure ingestion, entity extraction, graph write, index build, and one search request separately.
- Compare key settings such as chunk size, chunk overlap, worker count, batch size, and cache on/off.
- Write results to a stable JSON or Markdown report.
- Document expected hardware and provider assumptions.
Acceptance criteria
- The benchmark can run on a laptop-scale fixture without private data.
- Results include wall-clock time and enough configuration context to reproduce the run.
- Documentation explains which settings most affect runtime and where to change them.
- The profile complements existing support discussions about slow entity extraction instead of replacing them.
Motivation
Graph build and entity extraction latency are common concerns for new users. A small benchmark profile would make the speed-quality tradeoff easier to understand and would give maintainers a stable baseline for future optimization.
Scope
Acceptance criteria