Home Contact
Get Nivelato Log in
PhoneBench COIL
About

An AI-driven lab, on purpose.

Nivelato Labs is a software lab founded by Benj Lebron. It runs on a simple bet: a small lab that masters agentic development will out-ship a large team that does not.

The model

Agents build. A human directs.

Multi-agent development is the production line. Instead of one assistant answering one prompt at a time, the lab runs coordinated agents in parallel: one implements, one tests, one reviews, one documents. Work that used to queue up now compounds.

Direction stays human. Product calls, quality standards, and release decisions are made by a person who uses the software and answers for it. Agents propose and execute; the human disposes. Every release passes a human review, and every public claim traces back to a verifiable result.

Verification is a feature, not a chore. Agents check each other's work against source data before anything ships. Benchmarks like PhoneBench exist because the lab measures its own tools the same way it measures everyone else's: against a standard that does not bend.

The research loop is real too. PhoneBench, the lab's robotics AI benchmark, is autonomous agents research run on the lab's own methods: agents doing skilled physical work, scored in public. What the benchmark learns about agents feeds straight back into how the lab builds with them.

The output is real. Nivelato, the lab's offset measuring software, runs in production at real glass shops. PhoneBench publishes full transcripts, including the failures. The lab's credibility is its distribution: do the work in public, show the receipts, and let the results rank.

Mission

Software that measures up, for the trades that get ignored.

Nivelato Labs exists to build the software the big platforms never get around to: precise, field-ready tools for small shops doing skilled, physical work. Nivelato turns field measurements into quotes without the paper chase. PhoneBench keeps the lab honest about what its own methods can and cannot do.

The method is part of the mission. The work is built by coordinated AI agents, directed by a human who uses the software and answers for it, and verified against real results before it ships. If a product does not earn its place in the field, it does not ship.

Vision

A one-person lab that out-ships companies.

We see a near future where team size stops predicting software quality. A small lab with a directed agent fleet, honest benchmarks, and one accountable name on every release sets the pace: faster shipping, published evidence, and tools that respect the people who use them.

Nivelato is the proof of concept, already running in real shops. PhoneBench is the measuring stick, publishing what autonomous agents can actually do, failures included. The vision is not a bigger company. It is a better ratio of output to overhead, demonstrated in public.

Core values

What the lab holds to.

Four rules run every decision here, from what gets built to what gets published.

01 · MEASURE

Measure honestly

Numbers are the product. The lab benchmarks against standards that do not bend, and reports what the benchmark says, especially when it is unflattering.

02 · PUBLISH

Publish the failures

Transcripts, scorecards, failures and all. Credibility comes from receipts, not claims, so the work happens in public wherever it can.

03 · ANSWER

The builder supports it

The person who writes the software answers the ticket. No handoffs, and no account managers standing between a problem and the person who can fix it.

04 · STAY SMALL

Small on purpose

One accountable name on every release. Staying small keeps every standard personal: if it ships, a human here read it, tested it, and signed it.