Policy

Data Usage Policy

How ZeroWidth handles, processes, and protects the data you entrust to us — especially what we will and won't train AI on.

This policy explains how ZeroWidth handles, processes, and protects the data you entrust to us through our products. It sits alongside the Privacy Policy and the Data Processing Addendum; where those describe what we collect and the contractual frame, this one describes what we do with it once it's in our hands — especially what we will and will not use it for to train AI.

Our commitment

ZeroWidth provides tools designed to help businesses build, evaluate, and deploy AI-driven workflows. We are built on the understanding that the data you put into our tools — workflows you design in Workbench, evaluation rubrics and datasets in Caliper, knowledge pages in Compass, conversations in Chat — is significant and sensitive. We are committed to high standards of data integrity, control, and protection.

You own your data

You retain all rights, title, and interest in the data you upload to or generate within our products ("Customer Content"). This includes:

First-party proprietary content — workflows, datasets, prompts, rubrics, pages, conversations, ratings, agent configurations, integration credentials, and any other content you create.
Third-party data you are authorized to use within the services — content you have the rights to upload, store, and process through us.

ZeroWidth does not claim ownership of Customer Content. We hold the limited license described in the Terms of Service — only what is needed to operate the service for you.

We don't train on your content

ZeroWidth does not use Customer Content to train, fine-tune, or otherwise improve AI models — ours or anyone else's. Specifically:

The flows, prompts, datasets, evaluations, pages, and conversations you create are not used as training data for any model.
The inputs you send to AI providers through ZeroWidth and the outputs they return are not used as training data by us.
We do not share Customer Content with model providers for the purpose of model training.

There are two specific carve-outs from this, both narrow:

Limited moderation

A small number of authorized ZeroWidth personnel may inspect specific Customer Content when investigating a credible abuse report, security incident, or violation of the Acceptable Use Policy. This is the only routine internal-eyes-on-content scenario; we do not browse Customer Content for any other purpose.

Third-party AI providers

When you configure your flows to call a large language model — whether through our subprocessor OpenRouter or through a provider key you have configured directly (OpenAI, Anthropic, Google, etc.) — the relevant inputs and outputs transit to and from that provider. In every case:

The data is sent solely to deliver the functionality you have configured.
ZeroWidth does not repurpose this data for any other use.
The provider's own terms govern what that provider may do with the data on its side; you choose the providers, so you choose those terms.

We strongly recommend reviewing each provider's data-usage and training policies before configuring them. Many providers offer opt-out toggles for training-on-API-data; configure yours according to your organization's posture.

What we DO learn from

ZeroWidth does not use Customer Content to train models, but we may analyze and learn from the shape of how our products are used. This is where the distinction matters:

We do look at workflow-design patterns: the shapes of flows, the kinds of nodes that get wired together, how many steps a typical evaluation has, which integrations get added most often.
We do look at performance and operational metadata: latency, error rates, run durations, plan-limit thresholds, and similar.
We do use these signals (in aggregated, de-identified form) to improve our tools and to inform AI-assisted features that help you and other customers build more effective workflows.
We do not inspect the content of your flows, prompts, datasets, conversations, or other Customer Content for these purposes.

In short: we may learn from how the services are used. We do not learn from what you put into it.

If your organization has specific concerns about even aggregate, de-identified analysis, contact us — we'll work with you on the right scope under your master agreement.

Where your data lives

All Customer Content is stored on infrastructure located in the United States. Our products are served exclusively over HTTPS / TLS; any data transferred between you and ZeroWidth is encrypted in transit. Data is also encrypted at rest, with additional encryption for sensitive secrets such as integration credentials and API tokens.

For the full list of subprocessors that handle Customer Content on our behalf, and where each one operates, see the Subprocessors page.

How long we keep moderation records

When Customer Content is inspected as part of a moderation or abuse investigation (the "Limited moderation" carve-out above), the resulting case records are retained for as long as needed for the investigation and a short period thereafter, then automatically purged from our systems.

This is separate from the general retention shape that governs the Customer Content itself — see the Privacy Policy for the general retention shape across usage logs, account data, and Customer Content.

Your trust

If you have questions or concerns about how we handle your data — including data-usage practices, training scope, or moderation — please contact us. We treat trust as load-bearing.

contact@zerowidth.ai — subject line "Data Usage"

For data-protection-specific inquiries: dpo@zerowidth.ai

ZeroWidth, LLC