Preview  Clio is being built in the open. This page is the contract; it applies from day one.

Methodology

What Clio is, where the data comes from, and the rules we hold ourselves to — including where our own interests lie.

What this is

Clio is an English-language index of the embodied-AI industry: the models, the companies building them, the benchmarks they are measured on, and the papers that move the field. It is named for the muse of history — daughter of Mnemosyne, memory — because that is the job: keep the record, and keep it straight.

The reason it exists is a gap. A large share of the field's activity now happens in Chinese, and most of it reaches English-language readers late, second-hand, or not at all. Our team is based in China and reads those sources daily. Clio is where that reading is written down.

Where the data comes from

Three rules we don't bend

Position is never for sale.

No paid placement, no paid ranking, no pay-to-be-listed. Ordering comes from the data and the sort control you clicked. There is no arrangement under which money changes what you see here — and if that ever changes, this page changes first.

Every number carries its source.

If we cannot attribute a figure, we do not publish it. Where scores are involved, each one will carry both a source link and a provenance mark — official, self-reported, or reproduced — because those three things are not the same claim and should not look alike.

We declare our own interests.

Clio is built and paid for by Mnesis Labs, which has commercial interests in this industry — see below. Any entry involving Mnesis Labs, its products, its customers or its evaluation results is labelled as such wherever it appears.

Why there are no scores yet

The fastest way to look authoritative is to publish a leaderboard. The fastest way to stop being trustworthy is to publish one you can't source. Benchmark results are being assembled with attribution and provenance marks attached to each individual number, and they will ship when that work is done rather than when the page would look better for having them.

A second, separate track follows: Mnesis Arena, real-robot blind A/B evaluation run by Mnesis Labs — paired rollouts under matched conditions, ranked with confidence intervals instead of single-run point scores. Because we run it, it is the clearest case of rule three: it will be labelled as ours, its protocol published, and it will sit in its own track rather than being blended into aggregated third-party numbers.

Our interests, stated plainly

Mnesis Labs builds data infrastructure for embodied AI — capture, physical reconstruction, training and evaluation, deployed on customer sites. That means we are a participant in this industry, not a neutral observer of it. Several companies indexed here are potential customers, partners, or competitors.

Our answer to that is structural rather than promissory: Clio publishes facts with sources attached and no rankings of companies, the commercial site lives at a separate address, and nothing on these pages is a sales surface. If you find something that reads as a pitch rather than a fact, that is a bug — tell us and we will fix it.

Corrections

We will get things wrong: a mis-stated spec, a company in the wrong category, a claim attributed to the wrong lab, a summary that reads more confidently than the source supports. Send it to info@mnesislabs.ai and we will correct it. Corrections to published facts are made in the data, not quietly in the prose.

Scope, for now

Clio is in preview. The index is deliberately narrow at this stage — a curated set rather than an exhaustive one — because a small accurate index is more useful than a large unreliable one. It grows daily. English only: not because the Chinese-language audience doesn't matter, but because that audience already has these sources, and this side of the gap doesn't.