The agreement covering this site, Foundry and the contributor workspace, and the datasets, annotation, benchmarks, environments and embodied-data collections we deliver.
These terms are between you and Zyno AI Inc, trading as Perit. By using this site, Foundry or the contributor workspace you accept them. Where you have signed an order form, a master agreement, a data processing agreement or a Business Associate Agreement with us, that document governs and these terms fill the gaps.
You confirm that the voice or the person on camera is you, that you are over 18, and that you are recording where you may lawfully record and with the permission of whoever controls the premises. You grant Perit a worldwide, perpetual, transferable licence to use, reproduce, annotate and license your recordings, footage and their derivatives for building and evaluating AI systems.
Submissions are reviewed against the session brief. Approved work is paid at the rate shown on the session before you start — speech sessions from a published base rate per approved audio hour, with premiums for harder sessions and thinner locales. Rejected work is not paid and is deleted; where the fault was ours, it is paid anyway. Submitting audio or footage you did not record, synthetic material, or anyone recorded without their consent on the file ends your access and forfeits pending payment.
You are an independent contributor, not an employee. You choose which sessions to take and when.
You work to the written guide for the project, not your own preference, and you disclose any conflict with a company whose output you are asked to score. Nobody reaches paid production work before passing the training and testing modules, and work sampled below threshold routes back to calibration. Clips, images and text you see are confidential and stay inside Foundry — you stream them; you do not download them.
Freelance annotators are paid per approved unit at the rate on the listing. Fixed-term and full-time roles are governed by their own contracts, which take precedence over this section. Labels, rubrics and guides you produce are made for hire for Perit.
Applications are made through the form linked from the listing. We issue Foundry test credentials so you can complete the training and testing modules; those credentials are for you alone. We keep applications for up to twelve months after the role closes, then delete them unless you were hired.
Datasets are licensed, not sold. Unless your order form says otherwise, the licence is non-exclusive, perpetual and for internal model development and evaluation, and it covers training a model you go on to deploy commercially. You may not re-sell or redistribute the raw corpus, attempt to re-identify a speaker or a person in frame, enrol a voice or a face for identification or verification, or use footage to surveil or map a private premises. Exclusive and time-limited-exclusive terms exist and are priced on the order form.
Every engagement starts with a written spec and a pilot batch — five percent of the volume for collection and annotation, ten to thirty episodes for Physical Intelligence — delivered with a quality report. If the pilot shows the spec was wrong, we fix the spec and re-run the pilot before the main volume starts, and the re-run is not billed.
Material you send us to annotate is processed under your instructions as your processor, pinned to your region, streamed to reviewers rather than copied, and deleted on your schedule.
Benchmark results are provided as they are measured. We publish the method and the intervals; we do not warrant that a score predicts production performance on your traffic, and we will not adjust a run to improve a number.
Plan prices and unit rates are on the pricing page or your order form; physical-AI lines are priced from a pilot because the rig, environment access and operator training are scoped per program. Invoices are due 30 days from issue, in the currency stated, exclusive of tax. Work that fails QA is redone before delivery and is never billed to you. Contributor and annotator payouts are made after review, on the schedule shown in the workspace, by bank transfer or UPI depending on your country.
Perit owns the platforms, Foundry, the annotation guides, the rubrics, the benchmark design and the environments. You own your models and anything you train with a licensed dataset; nothing here transfers ownership of a trained model to Perit. The photographs and clips on this site that show capture settings are illustrative renders produced for the site — they are not customer data and are not samples of any delivered dataset.
Each side keeps the other's non-public information confidential. Our controls — encryption, role-scoped logged access, streamed review, region pinning, time-boxed handovers — are described on the trust page, and we hold SOC 2, ISO 27001:2022, ISO 27701:2019 and TPN Gold. Enterprise engagements include an annual audit right over residency controls and access logs on request.
We provide the platforms and the deliveries with reasonable skill and care and to the agreed spec. Beyond that, and to the extent the law allows, they are provided as they are. Neither side is liable for indirect or consequential loss, and our aggregate liability is capped at the fees paid in the twelve months before the claim. Nothing here limits liability for fraud, death or personal injury.
Either side may end an engagement in writing. We may suspend access immediately for a breach of acceptable use, for fraudulent submissions, or for footage that fails the consent rules. Licences already granted for delivered datasets survive termination; access to Foundry and the workspace does not.
These terms are governed by the law of the jurisdiction named in your order form, and by Delaware law where none is named, with exclusive jurisdiction in that place.
One address for privacy requests, security reports and anything contractual. We answer within 30 days, usually the same week.