Privacy Policy

1. Scope

This Privacy Policy ("Policy") is issued by Mojo AI Services LLC, a Florida limited liability company ("we", "us" or "our"). It applies to visitors to moonbridge.ai (the "Site"), people using staff accounts for a licensed therapy practice ("Authorized Users"), and clients enrolled by their practice ("Clients"). The Site and the Moonbridge application are the "Services".

Information we handle on a practice's behalf about its Clients ("Client Data") remains under that practice's control. Where HIPAA applies, the practice acts as the covered entity and we act as its business associate under a business associate agreement ("BAA"). This Policy explains our handling of information; it does not replace the practice's Notice of Privacy Practices, the BAA, or a Client's rights under applicable law. The BAA governs our processing of protected health information for the practice.

2. Information Collected Through the Site

The Site does not use advertising trackers, tracking pixels, or third-party behavioral analytics. We collect:

• Inquiries. Information entered in a practice registration or call-request form, such as name, work email, phone number, practice name, state, license or National Provider Identifier number, preferred contact time, and submission date.
• Connection and security information. Network address, browser information, requested pages, and information needed to detect abuse and keep the Site available. We use a limited-duration identifier derived from the submitting network address to limit abusive form submissions. We do not use connection information for advertising profiles.
• Display preference. A light or dark display choice may be saved in a preference cookie for up to one year. It is not used for advertising or tracking.

Cloudflare delivers and protects the Site and processes connection information. Practice inquiries are stored using Cloudflare services and may be emailed through Cloudflare to our Microsoft-hosted mailbox so we can respond. Public forms and general email are for business inquiries. Do not include Client Data, clinical records, or other health information in them.

3. Account and Service Information

We maintain the information needed to provide and protect accounts: names and contact details, practice details, staff roles, authentication and account-security information, accepted agreements and their editions, practice settings, and billing records. When online billing is turned on, payment card information is collected by our payment processor, not through a Client's record.

We record account and service activity, including sign-ins, actions affecting records, and permitted or denied access. For each sign-in we keep a device label, such as the browser and computer type, and a shortened form of the network address. We use them to tell the account holder about a sign-in from a new device and, for staff accounts, from a new network. We also process technical information needed to diagnose errors, measure service reliability, and investigate security incidents. If someone enables mobile notifications, we maintain the device delivery information needed to send them. Client Data is described separately in Section 6.

4. Use of Information

We use information to respond to inquiries; verify and set up practices; provide, maintain and support the Services; secure accounts and investigate misuse; deliver service and safety notifications; administer billing; and meet legal and contractual obligations.

Service reliability measurements use operational counts, timings and error codes rather than conversation content. Clinical information is used on the practice's behalf to provide the Services. Any separate use of Client content to evaluate or improve instructions, safeguards or clinical workflows, or for research, requires the applicable authorization and agreements. Accepting this Policy is not consent to such a program or permission to use Client content for general product development.

We do not sell personal information or Client Data, use it for advertising, or use it to train or fine-tune AI models. The same restrictions apply to service providers processing Client Data on our behalf.

5. Disclosure of Information

We disclose personal information only (a) to the service providers identified in Section 7, acting on our instructions and under written agreements; (b) as required by applicable law, regulation, subpoena, court order or other legal process; (c) to protect the rights, property or safety of Mojo AI Services LLC, the Services, our customers or the public, to the extent permitted by law; and (d) in connection with a merger, acquisition or sale of assets, subject to the commitments in this Policy. We do not disclose personal information to any other third party.

6. Client Data and Visibility

Clients join through their practice at the direction of their treating clinician. We handle the information the Client and authorized care team enter or generate in the Services, including account and consent information, companion conversations and memories, words saved for a session, messages with the practice, journal entries and private notes, check-ins and outcome measures, appointments, shared files, care plans, clinical notes, and safety-related records.

Private conversations and deliberate sharing. No clinician or practice staff member can read a Client's companion conversations or private companion memory, except the crisis-flag window described below. When a Client chooses Save for session, the app previews the selected words, any context the Client adds, and the named recipients before confirmation. Confirmation shares that version, not the rest of the conversation. A Client can change the words of a shared save. The change is shared as a new version with the recipients current at the time of the change, whom the app shows before the Client confirms it, and copies already received are not taken back. A shared save cannot be removed or emptied. The rest of the conversation is still reviewed by AI to prepare the clinical observations described next. A change in care team does not automatically give new recipients access to earlier shared versions. Direct messages, check-ins, shared files and clinician-authored records follow their own disclosed care-team permissions.

Clinical observations. Under the Client's current companion sharing agreement, a separate AI model reviews the Client's companion conversations, the Client's messages and the companion's replies, including messages the Client did not save, and prepares bounded clinical observations and possible explanations. The observations go to the clinicians and supervisors on the care team whom the agreement names, and reading them requires a current care relationship. These can include sensitive inferences, uncertainty and alternative explanations. They are for clinician review, not automatic diagnoses, patient reports or a transcript-browsing route. The clinician remains responsible for assessment and treatment decisions. Accepting the companion sharing agreement includes this analysis. A Client who does not accept it does not receive companion replies. Once a Client has accepted, there is no setting in the app that turns the analysis or the companion off. A Client who wants to stop can stop using the companion, ask the practice to remove companion access, or close their own account in the app. Removing access or closing the account stops new companion replies and future analysis. Neither erases observations, saves or other records already kept.

Clinician AI tools. Clinicians can use AI tools built on the same model. A drafting tool prepares an editable brief, plan or formulation from observations already released to them and material the Client confirmed. A writing tool turns a clinician's own notes into an editable note, summary, plan or between-session activity. A clinician chat answers the clinician's questions from the Client's chart: plans, notes, the safety plan, between-session activities, outcome measures, words the Client saved for a session, and summaries of crisis flags. None of these tools reads the Client's full companion conversation or private memory. Their output is a draft for the clinician, who decides what, if anything, becomes part of the record or the plan.

Concern review and urgent safety. A Client can send a concern about the companion to the clinician for review; the confirmed report and selected context are retained, and a notification is queued for the clinician. Delivery may fail. Other enabled supervision or safety processes have their own explained purpose, recipients and limited evidence. A concern does not itself authorize access to a full conversation. A crisis flag is the safety screen's highest tier, raised when a message suggests someone may be in danger now. Only a crisis flag reaches practice staff. The treating, supervising and covering clinicians, and an owner who is a clinician, see the flagged message with up to five messages before it and two after it. No other flag is shown to staff. Administrative or front-desk authority alone does not grant that clinical evidence. Automated notices do not establish that a clinician has read or responded.

Private writing and files. Journal entries and private notes are the Client's alone. No clinician, practice staff member or AI reads them, and they raise no flag. If the Client dictates, Amazon's speech service turns their voice into text. Crisis help appears on the device to the Client only. A Client may seal them on the device with a passcode of at least four characters, which cannot be recovered if forgotten. Once unlocked, they lock again after 15 minutes without a tap or key press in the app. Sealed, we hold only text that cannot be read without the passcode. Someone who obtains the sealed text could guess a short passcode; a longer one is much harder to guess. Unsealed, they are stored by us, and a court could order them produced. Backups keep earlier and deleted text, including text written before it was sealed, for up to 35 days. Unshared files have separate access controls. A private label does not establish legal privilege.

AI and dictation. The companion and clinician-support features are built on Claude, an AI model developed by Anthropic, which we run through Amazon Bedrock, an AWS service, inside our own AWS account in United States regions, with Moonbridge's instructions and safety controls. To produce a reply, we send the message a Client writes, the recent conversation, any companion memories the Client keeps, and the clinician-approved care context, such as the treatment plan and the safety plan. Journal entries and private notes are never sent to the AI. The same service runs a second reading of each companion message for risk, alongside the safety screen, as well as the clinical observation review and the clinician AI tools described above. AWS processes that content only to return the response; it is not transmitted to Anthropic and is not used to train any model. A Client agrees to this processing in the app before the companion sends anything. It is software, not a person or a licensed professional, and its output can be wrong. If a Client or staff member chooses dictation, Amazon Transcribe, an AWS service, processes live audio and returns text to the box they are speaking into. Moonbridge does not keep the recording as an audio file. Our AWS account is opted out of AWS AI services using its content to develop or improve their services, so dictated audio is not used that way. A dictated message is sent to the companion only when the Client sends it.

Spoken replies. Spoken replies, in which the companion reads its replies aloud, are optional. When a Client turns them on, the text of a companion reply is sent to Amazon Polly, an AWS service, within our own AWS account in United States regions and under AWS's business associate addendum. Polly returns synthetic audio. Moonbridge does not save that audio, and no recording of anyone's voice is made. The Client can turn spoken replies off at any time. Spoken replies come with Talk, in which Amazon Transcribe turns what the Client says into text, as it does for dictation. Unlike a dictated message, that text is sent to the companion when the Client stops talking for a moment.

Storage and operation. Application records and files are hosted in our AWS environment in United States regions, with encryption in transit and at rest. Application traffic also passes through Cloudflare as described in Section 7. Access by our personnel is restricted to authorized service, support, security, and legal purposes and is subject to access controls and audit procedures. Our operations portal does not show companion conversations, journal entries, private notes or clinical records.

Records and rights. The practice controls its clinical record and retention obligations. Clients may request access, correction or a copy through the app's records process or directly from the practice; we assist as the BAA requires. A clinician-only screen does not remove applicable record-access rights. The practice evaluates the requested records and any lawful exceptions; information used to make decisions about an individual is not excluded merely because an AI system helped prepare it. Corrections are linked to retained records rather than silently rewriting signed history. Legal protections and disclosure duties depend on applicable law and the circumstances. A court or another lawful process may require records to be produced, including unsealed private writing. For sealed writing we hold only text that cannot be read without the passcode.

7. Service Providers

We use service providers for the functions below, subject to applicable contracts and data-protection obligations. The BAA each practice signs lists the providers that process protected health information on our behalf and states which of them a business associate agreement covers. A provider's certification does not certify Moonbridge.

• Amazon Web Services (AWS). Hosts application services, databases, files and backups, and provides AI processing, dictation transcription, and email delivery. Our clinical storage and AI processing are configured for United States regions. Client content is processed to deliver the requested service under the applicable AWS terms and our data-use restrictions. Amazon Bedrock, the AI model service, does not use that content to train models, and our AWS account is opted out of AWS AI services, Amazon Transcribe included, using its content to develop or improve their services. We do not represent that all provider operational or security records have zero retention.
• Anthropic. Developer of the Claude model behind the companion, the clinical observation review and the clinician AI tools. We run the model through Amazon Bedrock inside our own AWS account, so Client content is processed by AWS under the terms above and is not transmitted to Anthropic. No account, billing or contact information is shared with Anthropic.
• Cloudflare. Delivers and protects the Site and application and processes application requests and responses on their way to and from our application services. That traffic can contain Client Data. Cloudflare also processes connection and security information. It is not the application's clinical-record database. Its role is not limited to serving static files.
• Stripe. When online billing is turned on, Stripe processes practice subscriptions, billing contact details, payment information and billing references. Until then, our team records companion licenses by hand, and no payment card is collected through the Services. We do not send Client identities, conversations or clinical records in our billing integration.
• Notification delivery. Amazon Simple Email Service, an AWS service, sends email. Where phone notifications are turned on for staff, Apple's push notification service delivers them. Staff receive notices including sign-in alerts, crisis flags, review requests, record updates and a daily notice of new messages. A Client may receive email about setting up and getting back into their account, a change to their password or email address, a sign-in from a new device, a session the practice canceled, moved or restored (when the clinician chooses to send it), and their records requests. These messages carry no client names and none of the words a client wrote. A crisis notice says only that a crisis flag is waiting and, where it applies, that the flagged wording reads like an intrusive thought. A session notice carries no date or time. A sign-in notice names the device and the practice. Anyone who can open the recipient's mailbox can read them.
• Business inquiries. Cloudflare and Microsoft process public inquiry information as described in Section 2. These channels are not for Client Data.

The current service-provider and agreement review is available to practices on request. We assess a provider's role based on the information it actually handles, including information in transit.

8. Retention and Export

We retain information for the purposes described in this Policy and as required by applicable law and our agreements. Business inquiries are retained while being handled and under our inquiry-retention process. Account, billing, security, and audit records may need to remain after an account is deactivated.

Client Data follows the practice's retention policy and BAA. A billing lapse does not remove access to records, export, or safety paths. Termination is handled under the practice agreement and BAA, with data return or export and any required retention or destruction. We do not promise automatic destruction of all records after a fixed period following termination.

The practice may set a retention ceiling for raw conversation text, private journal text and original files that have not been shared. Where no ceiling is configured, there is no scheduled destruction under that setting. Records subject to a legal hold are preserved. Deleting an item from an active view does not mean that required records or backup copies are immediately erased.

9. Individual Rights

An individual may request access to, correction of, or deletion of personal information that Mojo AI Services LLC holds about that individual by writing to info@moonbridge.ai. We respond within thirty days. We honor deletion requests except where retention is required by applicable law, by a legal hold, or by the agreement with a practice, in which case we inform the requester of the basis for retention. Requests by a client concerning the client's clinical record are directed to the client's practice, which controls that record; Mojo AI Services LLC assists the practice in responding as required by the business associate agreement.

10. Security and Incident Response

We maintain a written security program addressing administrative, physical and technical safeguards, including access controls, multi-factor authentication, encryption, audit records, incident response, backup and recovery, and vendor review. Appropriate program and assessment materials are available to practices on request. These safeguards reduce risk but do not guarantee that every incident can be prevented.

For protected health information, our incident and breach duties follow applicable law and the BAA. We notify the affected practice without unreasonable delay. For a breach of unsecured protected health information, we notify no later than five business days after discovery, as the BAA each practice owner signs in the app provides. We assist with the practice's response and notification obligations. This Policy does not claim a HIPAA certification or a completed Moonbridge SOC 2 examination.

11. Age and Eligibility

The current Client enrollment program is for adults eighteen or older and is available only where the Services permit enrollment. The practice determines eligibility and obtains the required consent. Public inquiry forms are intended for practice representatives, not children or Clients. Any future adolescent program requires its own eligibility and consent arrangements and an updated disclosure before enrollment.

12. Changes to This Policy

We may amend this Policy from time to time. Each edition carries a version number and an effective date, shown on the website. A new edition applies from its effective date. The app shows it to each Authorized User, who must agree to it before continuing, and to each Client, who must agree to it before using the companion again. When a change is material, we also tell each practice's owners by email. Prior versions are available on request.

13. Contact

Mojo AI Services LLC, 150 East Palmetto Park Road, Suite 800, Boca Raton, Florida 33432, United States. info@moonbridge.ai.

Version 2.1, in effect from October 5, 2026. Questions regarding this document may be directed to info@moonbridge.ai.