Privacy Policy

Effective date: 2026-09-23

Protostar AI, LLC (“Protostar”, “we”, “us”) operates this website at protostarai.com and the Protostar Secure AI Gateway at api.protostarai.com. This policy explains what data we collect, what we do with it, who else is involved, and what choices you have. It is written to describe the systems as they are actually built. Where a practice is a design goal rather than a certified fact, we say so.

Protostar operates in the United States only. Contact: hello@protostarai.com.

1. This website

The website collects nothing except what you send through the contact form. protostarai.com is a static site: no accounts, no cookies, no analytics or advertising trackers, no third-party scripts, and no access logs.

If you use the contact form, your name, email address, and message are sent to us by email and kept only in our mailbox; the site itself stores nothing. A single function hosted on Amazon Web Services relays the message and records only an anonymous request identifier and whether the send succeeded. We use what you send solely to reply to you. If you email us directly, the same applies.

The site is hosted on Amazon Web Services (S3 and CloudFront). AWS may process your IP address transiently to serve the page, as any host does.

2. The Protostar Secure AI Gateway

The gateway is a business service. Our customers (“tenants”) connect their own applications to it so that their users can use AI models without exposing sensitive data. Two kinds of data are involved, and they are treated very differently.

2.1 Content you send for processing

This is the text your application sends to the gateway: prompts, documents, chat messages. It may contain personal information, including protected health information, depending on what your application does.

• We process it; we do not keep it. Content is handled in memory for the duration of a single request and is never written to a database, a log file, or any persistent store.
• We never use it to train or improve any model.
• Detected identifiers are held in a request-scoped vault inside the gateway and are returned only to your authenticated application. They do not appear in our records.

2.2 Records we keep about your tenant

To run the service, bill it, and prove what it did, we retain:

• API keys, stored only as a one-way hash. The raw key is shown once at creation and we cannot recover it.
• Tenant and organization identifiers and the contact details on your account.
• Usage metering: request counts, estimated token counts, and estimated cost, by day, tier, and data class.
• An audit record for every request, containing the tenant, data class, routing tier, whether the request left the boundary, the destination provider if it did, the number of identifiers masked, token and cost estimates, and timestamps. Audit records contain no message content, no identifiers, and no real values. They are chained with cryptographic hashes so that tampering is detectable.
• Provider keys you bring yourself (if you use that feature), encrypted at rest, never displayed after entry, and never written to the audit.

2.3 How AI is used, and where your content goes

The gateway is an AI service, and this is the disclosure that matters most.

Every request is classified by sensitivity into one of three classes. Your application can declare a class, and a declaration can only make the classification stricter, never looser.

• Class C1 (regulated-sensitive, including health and financial data) is processed only on a model we host ourselves, IBM Granite, inside a private network with no route to the internet. C1 content never leaves that boundary. This is enforced by the network design, not by policy.
• Classes C2 and C3 (confidential and low-sensitivity) may be sent to an external model, currently Anthropic’s Claude, reached through Amazon Bedrock in our own AWS account. Before any content leaves, detected personal identifiers are replaced with realistic stand-ins. The real values stay inside the boundary and are restored into the answer after it returns. The external model never receives them.

Before content is sent, personal identifiers are detected by a pattern layer and a named-entity recognition layer. Detection is designed to favour recall, but no detection system is perfect; the guarantee that regulated data stays inside the boundary comes from the C1 routing rule, not from detection alone. Customers handling regulated data should declare class C1.

Outbound content may additionally be rewritten by our self-hosted model to reduce stylistic fingerprints before it leaves. This rewriting happens on the anonymized text only.

2.4 Third parties

We use exactly these third parties in the gateway:

• Amazon Web Services hosts the gateway, the self-hosted model, the database, and the secrets store, and provides Amazon Bedrock. AWS has signed a Business Associate Agreement with Protostar covering this account. AWS states that Amazon Bedrock does not use customer inputs or outputs to train its models.
• Anthropic, whose Claude models we reach only through Amazon Bedrock and only with anonymized C2 and C3 content. We have no direct data relationship with Anthropic.

We do not sell personal information. We do not share it with advertisers, data brokers, or analytics providers. There are no other processors.

2.5 Retention and deletion

• Content: not retained at all (see 2.1).
• API keys and provider keys: until you revoke them, which you can do yourself at any time from the console or the API.
• Usage and audit records: retained for the duration of the service relationship to support billing and compliance. Audit records are removed only by an explicit operator action, never automatically, so that the record stays complete for as long as the relationship lasts.
• Deletion requests: to have your tenant records deleted, email hello@protostarai.com from your account contact address. We will acknowledge your request and tell you what will be deleted and what must be kept (for example, to preserve the integrity of the audit record or to meet a legal obligation) before we act.

Because audit records are hash-chained, deletion of old records is performed by sealing and removing the expired portion of the chain, which preserves the integrity of what remains.

3. Security

Data in transit is protected with TLS. The database is encrypted at rest. The gateway and the self-hosted model run in private subnets with no internet gateway and no NAT, reaching AWS services only over private endpoints. Secrets are held in AWS Secrets Manager and never in code.

Protostar is designed to HIPAA and SOC 2 controls. We have not been audited or certified against any standard, and we do not claim compliance with one. Before any protected health information is processed for a customer, a separate Business Associate Agreement between that customer and Protostar is required.

4. Your rights and choices

Depending on where you live, you may have rights to access, correct, delete, or restrict the processing of your personal information. Requests can be sent to hello@protostarai.com. If you are an end user of a customer’s application rather than a customer yourself, the customer controls your data and your request should go to them; we will assist them in fulfilling it.

We do not use automated decision-making that produces legal or similarly significant effects about individuals.

5. Children

Our services are business tools and are not directed to anyone under 18. We do not knowingly collect personal information from children.

6. Changes to this policy

When we change this policy we will update the effective date at the top of this page. Material changes affecting gateway customers will also be sent to the account contact.

7. Contact

Protostar AI, LLC hello@protostarai.com