Get Started · Benchmark

Don't take our word for it.
Run the benchmark yourself.

DocumentForge — the open Order Database — ships a real stress benchmark on airline-shaped data. Hand the prompt below to your AI agent, or run one command yourself. No account, no sales call.

0.01 ms

Indexed PNR point lookup (avg)

~85K

QPS sustained, 100k-doc quick run

~40 s

Quick benchmark, end to end

External dependencies to install

Point an agent at it

One prompt. A few ways to run it.

The quickest path is to paste the prompt into your coding agent and let it clone, build, and read the numbers back to you. Prefer to drive? Pick a variation in the terminal — clone & build with the .NET 9 SDK, or download the prebuilt dfdb binary and skip the toolchain entirely.

Paste this to your AI agent

Works in Claude Code, Claude.ai, Cursor, or any coding agent — it will clone the repo, run the benchmark, and read back the numbers.

Clone the DocumentForge repo and run its benchmark, then summarise the results for me.

Repo: https://github.com/aerotoysio/documentforge
Quick mode is ~40s on 100k airline-order documents:

  git clone https://github.com/aerotoysio/documentforge.git
  cd documentforge
  DFDB_BENCH_QUICK=1 dotnet run -c Release --project samples/DocumentForge.Benchmark

You'll need the .NET 9 SDK. When it finishes, show me the QPS + latency
table it prints and tell me whether the indexed point lookup really is
sub-millisecond.

documentforge — benchmark

# ~40s · 100k docs · needs the .NET 9 SDK
git clone https://github.com/aerotoysio/documentforge.git
cd documentforge
DFDB_BENCH_QUICK=1 dotnet run -c Release \
  --project samples/DocumentForge.Benchmark

Browse the source Benchmark report Methodology & docs

What you'll see

The same workload an airline runs.

The benchmark bulk-inserts airline-order documents (PNRs, passengers, nested flight segments), builds persistent indexes, reopens cold, then sustains each query type for a fixed window — reporting average, min, max latency and QPS per query.

Indexed point lookups

PNR equality against a unique index — the hot path for order retrieval. Sub-millisecond, tens of thousands of QPS.

Selective & nested scans

Status filters and nested array paths (departure airport inside flight segments) over indexes.

Cold reopen

Indexes are loaded from disk on restart — no rebuild — so the second start is as fast as the first was slow.

About the Order Database