Performance engineering as a service

What it is

Breakage is rude: it fails a test, pages somebody, and a person has to stop. Performance is much more polite. A hot path gets ten percent slower and the suite stays green, data grows, and six months later everybody agrees the system is slow, but no single change ever demanded a meeting. Performance engineering as a service means the work that catches that quiet failure, delivered continuously, without hiring for it. Perfloop learns your system from code and telemetry, and hunts a wide catalog of performance patterns. A change reaches you only after your checks pass, the speedup is measured against the baseline, and disproving it has failed. What you get is a stream of proven pull requests, and software that stays fast. What fails is recorded as refuted, so nobody spends time on it again.

In four nouns: a model of your system, a catalog of performance patterns, a harness that tries to disprove every change before it reaches you, and controls that decide what runs on its own.

What you get

• A living map of your system: components, hot paths, workloads, and the performance hypotheses anchored on each.
• Proven pull requests. Each one carries its measurements and the attempt to disprove it. Proof opens the PR; the merge stays yours.
• A backlog that works itself. Your engineers' own ideas can be submitted as hypotheses and proven the same way.
• A record of what did not work, which is worth as much as what did.

How it differs from a consultancy

A consultancy bills hours and leaves. The findings live in a report, and the report ages the day your code changes. Perfloop runs continuously against the code as it is today. Keeping a system fast was never a task you finished. The performance envelope moves when a dependency changes, when an agent builds today's feature on top of the one it wrote yesterday, and even when the source stands still. A practice that visits once a quarter cannot hold an envelope that moves every week.

How it differs from a profiler or an APM

Profilers and APMs show you where time goes. They do not propose the change, build the experiment, or prove the result, and a benchmark suite in CI only asks the questions somebody already thought to encode. Perfloop looks for the missing question, builds the experiment, and, when production telemetry is connected and a change declared a post-deploy telemetry check, checks the result once more after the merge. The observability you already have becomes an input, not a replacement.

How it differs from asking a coding agent

Let a coding agent run free and the bill moves to the reviewer, who receives a plausible diff and must work out whether behavior changed, whether the workload was fair, and whether the number survives a clean baseline. Steer it closely and the engineer pays the same bill one prompt at a time. Neither arrangement removes the proving work. It only chooses the person who gets stuck with it. Perfloop turns the lab into the product: a case asks one question about one place in the code, the project's checks run, the agent that wrote the change does not grade it, and only a candidate with a clean verification reaches the comparison, where the original and candidate are measured under the same rules in the same environment.

If you already have performance engineers

Then this is for them. Perfloop is a catalyst the human steers. The machine writes the benchmark, reruns the baseline, throws out the miraculous result, and comes back when the workload changes; it can be patient where people cannot afford to be. Your engineers stay on the judgment the machine cannot make: which tradeoffs, what is worth holding, whether a change belongs. Not steering more, steering higher. Their own hunches go in as hypotheses and come back proven or refuted, which gives the hunch more reach. Compute goes on repeatable proof and expert attention on judgment, and the path from a suspicion to a validated improvement gets short enough that more real performance work becomes worth doing. Let the machine spend the thousand hours. Give the engineer back the afternoon.

Trust

The agent that wrote the change does not grade it. A separate verifier looks for defects before any performance result can anchor the review, and a speedup is a claim about real hardware under a realistic workload, settled with samples and controls. Perfloop can repeat the experiment later, and, when production telemetry is connected and a change declared a post-deploy telemetry check, check the result once more after the merge. Perfloop cannot merge, approve, or deploy. The same machine runs in public: our optimizations have been merged upstream into ClickHouse, Prometheus, ruff, parquet-go, RoaringBitmap, zoekt, and tag. Every case and its evidence is on the public instance.

Pricing

Pro is $5,000/month, month-to-month; Enterprise from $15,000/month, annual. The Evaluation is $0 for seven days, by invite, on your own code. The anchor is a performance engineer's salary, not a developer tool.