Writes the script
write_file and edit_file create and adjust your benchmark script, and every change appears as a diff.
Product
Server benchmark scripts, runs and logs for machines you own
01
A server benchmark is a fixed piece of work you run on a machine you own or may test, with the same settings each time, so two results compare fairly. BenchQuorum writes the script, runs it in the shell and records only the numbers that were measured. Scope stays with you.
Runs your command
Shell runs your test command and reports the real exit code, so you can see when a run failed.
Reads the numbers
read_file and grep collect numbers from saved result files, and the agent builds a table of runs.
What the agent can do
- Filesread_file, write_file, append_file, edit_file, delete_file, make_dir, move_file, copy_file, list_dir, grep and glob work inside the chat's working folder, with a counter of files created, edited and read.
- Shellruns terminal commands such as your test command or systeminfo, and judges success by the real exit code. Dangerous commands are blocked.
- Webweb_search finds pages, and http_get reads one documentation page, for example the manual of your own hardware.
- Screenshots and imagesocr reads text from a screenshot, such as a settings screen or a result window, and gen_image saves real image files.
- Orchestrationtask starts parallel sub-agents for separate checks, run_on_model sends a hard step to a bigger model, and skill stores reusable playbooks.
- Approval modesAuto-run acts immediately, Ask first pauses before every action that changes something, and Plan only plans and changes nothing. Plans survive restarts.
- Cronruns a prompt on a schedule you set, for example a weekly run that appends one row to your results file.
- Modelsconnect a cloud AI provider through a standard API endpoint, a model running locally on your PC, or the built-in model.
Where the work happens
The app runs on your Windows PC inside one working folder. A command that reaches another machine is one you asked for, on a machine you own or may test.
Write bench.py, run it on this machine and save the median with the date to results.csv.
- write_file
bench.pysaved - Shell
systeminfoexit 0 - write_file
system.txtsaved - Shell
python bench.pyexit 0 - append_file
results.csvrow added - read_file
results.csvread - edit_file
notes.mddiff shown
Waiting for your OK before the next stepRejectAccept
Two tasks, step by step
Set a baseline for a homelab machine
Before tuning a homelab machine for benchmarking, take a baseline: ask the agent to run your benchmark script on the untouched machine first. Shell runs it and reports the exit code, write_file saves the output as baseline.txt, and append_file adds the median to results.csv. Change one setting, then ask for the same run again and the agent shows both rows side by side. One change at a time. Read the plan in Plan only mode before anything runs, then switch to Ask first so every file write waits for your OK, and compare the medians yourself before you keep the change.
Check a rented VPS from your own PC
To benchmark a VPS you rent, the disk and the network need separate tests, because each measures a different part of the machine. Check the provider's terms for testing first. From your PC, Shell runs ping -n 20 followed by your VPS address, and write_file saves the round-trip times as latency.txt. You run the disk test yourself on the VPS, outside BenchQuorum, with your own command. VPS network and CPU test results only mean something next to the plan, the time of day and the settings you recorded. Keep the traffic small and short, and check each number against the console output.
What BenchQuorum does not do
BenchQuorum is not a hosted service or a plugin for another program. It is meant only for your own files, your PC and machines you are authorised to use, so do not point it at systems, accounts or data you do not own. It cannot guarantee that code, measurements or analysis are correct, so you review every diff.
