Perfloop
1 known investors
Perfloop is an automated performance-engineering system that finds, tests and proves code speedups, delivering them as pull requests.
Also known as perfloop.ai
Investors · 1
Company profile
researched Aug 2026Perfloop builds an automated performance-engineering system for software teams. The product ingests a customer's code and telemetry to construct and continuously update a model of the system, then searches that model against a catalog of performance patterns — examples cited include inefficient algorithms, allocation hot loops, data-parallel gaps and synchronous waits. Candidate improvements are handled as "cases": each hypothesis is worked, measured against a baseline, and subjected to attempts to disprove the claimed speedup; only changes that survive the customer's own checks and the measured comparison are surfaced as pull requests, with merge authority retained by the customer's engineers.
Usage is configured per repository through four switches — model upkeep (on by default, rebuilding the system model as code changes), case work (opt-in triage of the case frontier), PR feedback (on by default, resuming a case when maintainers comment on an open pull request) and pull requests (opt-in, opening a proven PR when a case holds) — plus a research cadence in which the user sets a goal and the system tests new findings from the field against their code, cheapest test first; accepted hypotheses become cases. Founder Tomás Senart frames the product as closing the second half of the measurement loop: after measurement, the system finds performance opportunities, proposes changes and proves whether they moved the needle, while human engineers decide what ships.
Founding story
Perfloop was founded by Tomás Senart, an open-source builder who created the Go-based HTTP load-testing tool Vegeta while an engineer at SoundCloud, where he needed a load-testing tool that avoided coordinated omission and the limitations of then-available options such as Apache Benchmark. He carried lessons from that work — in particular that a benchmark run once is already stale and that performance must be re-verified against constant change — into the company.
Business model
Software product delivered against customer code repositories, with access currently by early access and self-serve access stated as planned for later in the year [0].
Traction
As of a 2026-08-24 snapshot published by the company, its model covered 129 repositories, with 333 cases run, 75 validated across 40 repositories, 23 merged upstream and 3,286 measured patch revisions. The company's GitHub organization hosted 131 repositories, largely forks of open-source projects including roaring, VictoriaMetrics, ClickHouse, prometheus, neon, go-ethereum, parseable, uint256, codex and casei, with commit activity recorded through August 2026.
Latest developments
A founder interview published in July 2026 detailed the product's approach to automated benchmarking and proof of speedups. The company published a metrics snapshot dated 2026-08-24, and its GitHub organization showed repository updates through late August 2026.
▸Full profile — market position, technology, go-to-market, geography
Market position
Positioned in performance engineering for software teams, with the founder arguing the need is growing as AI coding agents write a larger share of code and introduce more regressions.
The company emphasizes verification: a proposed change is surfaced only after the customer's checks pass, the speedup is measured against a baseline and attempts to disprove it fail, with merge decisions left to the customer's maintainers.
Technology
Perfloop learns a model of a customer's system from code and telemetry and keeps it current as the code changes, then hunts a catalog of performance patterns. Each candidate change is measured against a baseline and must survive attempts to disprove the improvement plus the customer's existing checks before a pull request is opened. Public demonstration work has been run against open-source projects including Zoekt, ClickHouse, Prometheus, RoaringBitmap, Tigris TAG, Ruff and parquet-go.
Go-to-market
Access via an early-access sign-up on the company website, with self-serve availability stated as coming later in the year; visibility is also built through performance work on well-known open-source projects, some of it merged upstream, and through founder commentary on performance engineering.
Software engineering teams maintaining codebases where latency and infrastructure costs are material; the company is described as serving fast-growing AI-native teams.
Geography
Investor geography on record is the United States.
Compiled by commissioned research from 7 cited public sources — announcements, filings, and press listed under research sources below.
Key figures
latest reportedCompany-reported or press-reported figures, each dated to when it was claimed — not independently audited.
▸Research sources · 7
primary sources listed
- Perfloopperfloop.ai · web
7 public sources were cited for this profile; the first-party ones are listed here.
Frequently asked questions
- What does Perfloop do?
- Perfloop is an automated performance-engineering system that finds, tests and proves code speedups, delivering them as pull requests.
- Who are Perfloop's investors?
- Perfloop's investors include Mango Capital.