Find your RAG bottlenecks before tuning under load.
What it does
Benchmarks a deployed RAG server and produces a unified report of stage timings, bottlenecks, citation quality, and optional load-test performance.
When to use it
Use for retrieval or reranker tuning, performance regression checks, and concurrency or retrieval-setting sweeps.
How to use it
Give it a benchmarking request in the repository checkout; it customizes a YAML preset, runs the benchmark, writes reports, and summarizes the results.
What you provide
- An existing project
Uses
NR
Latest review
What you need · 4
Requires Python 3.11 or newer.
Requires uv on PATH to install dependencies and run the benchmark from the repository root.
Access · 3
This skill
RAG server
Write
Sends benchmark settings to RAG
ingestor server
Read
Checks ingestor collections