Privacy Policy.

1.1. No Real Names Required. We do not track your real name, and frankly, we do not want it. When you create an account, we ask you to pick a safe nickname. To our systems, you are just "StarGirl" or "Bluebird."

1.2. Identity Separation. The system architecture is built like an iron wall. We store your account credentials in one locked vault, and your health conversations in a completely different one. Even if our database was somehow compromised, the attackers could not tie your specific health questions back to your identity.

We believe in strictly minimal data collection. We only hoard data if it directly benefits your experience. Here is exactly what we keep:

• Your Email Address: Collected solely so you can reset your password. We do not use it to track you across the internet.
• Your Chat Text: Saved so you can read your past advice and pick up conversations where you left off.
• General Region: We need to know roughly where you are so that if you are in severe danger, the AI gives you the correct local emergency numbers (e.g., 112 in Nigeria vs. 911 in the US).

4.1. Disabled by Design. Normally, privacy laws mandate a feature called "Data Portability"—the right to click a button and download all your data. We have intentionally disabled this for your chat logs.

4.2. Security Over Convenience. Why? Because these logs are retained strictly to train our AI to be smarter; they are not meant to be a personal scrapbook. Leaving a PDF titled My_Deepest_Insecurities.pdf lying around in your local downloads folder is terrible op-sec. We keep your secrets locked down in our encrypted vaults, for the system's eyes only.

5.1. Scanning for Predators. Our platform features a "Red Flag Detector" where you can paste screenshots or text from other apps (like WhatsApp or Instagram) to check for grooming, gaslighting, or manipulation tactics.

5.2. Ephemeral Processing. When you paste these external chats, the AI analyzes the text, identifies the toxic behavior, generates a safe reply, and then immediately purges the pasted text from our active memory.

5.3. We Don't Save Trash. To put it plainly: if a guy named Chad is trying to manipulate you, we will analyze his tactics to protect you, but we do not permanently archive Chad's emotional abuse in our databases. It is processed and dumped.

6.1. Extreme Constraints. For our child-focused AI Buddy modules, data processing operates under extreme NDPR and COPPA constraints. The AI acts as a digital shield, mentor, and educational guide.

6.2. Cryptographic Parental Consent. We do not build marketing profiles on children. Period. Before a minor can initiate their first chat, parental or guardian consent must be cryptographically logged and verified through our signature matrix (found at the bottom of this document).

7.1. Safe Exploration. We provide specific modules for young adults to help them understand their rights, set personal boundaries, and navigate complex social dynamics. Conversations in these modules are highly sensitive.

7.2. Elevated Privacy Routing. Queries routed through the young adult boundary modules are assigned the highest level of contextual anonymity. The system is designed to provide actionable advice without attaching the vulnerability to a permanent, identifiable user record.

8.1. Weaponization Defense. Information regarding your reproductive health and menstrual cycles is classified as hyper-sensitive data. We encrypt this data at rest using AES-256 and do not allow it to be co-mingled with general demographic analytics.

8.2. Absolute Control. You maintain absolute control over your cycle data. It is never used to target you with advertisements for health products, and it cannot be queried by unauthorized internal staff.

10.1. Encryption Matrix. As soon as you hit send, your message is wrapped in an AES-256 digital lock. It travels through the internet via TLS 1.3 secure tunnels. At rest in our databases, the storage volumes are fully encrypted.

10.2. Key Management. The decryption keys are physically separated from the database clusters. Without the proper application-layer authentication, a stolen hard drive looks like complete mathematical gibberish.

11.1. Who Touches the Servers? We build the software, but we don't own the physical server racks. We use premium, audited third-party infrastructure (like Aiven Valkey and secure model hosting providers) to run our systems.

11.2. Zero-Knowledge Processing. These infrastructure partners process the data, but they do not have the keys to read it. They simply provide the computing power. We hold them to strict Data Processing Agreements (DPAs) that legally bind them to our privacy standards.

12.1. Legal Compliance. If we receive a legally valid, court-ordered subpoena, we are obligated to comply with the law. We do not operate above it.

12.2. The Reality of the Iron Wall. However, because your real identity is decoupled from your encrypted health data, handing over a file that says "Anonymous User #8274 asked about menstrual cramps" is not exactly the smoking gun authorities usually hope for. We can only hand over what we have, and we design our systems to have very little identifying information.

13.1. Safety Over Output. We implement strict constitutional AI guardrails. If a prompt attempts to bypass safety protocols or requests instructions for self-harm or violence, the model intercepts it and refuses to generate a harmful response.

13.2. Bias Mitigation. Because the AI is trained on historical data, we actively monitor and patch the models to prevent racial, geographic, or gender-based medical biases from creeping into the advice.

15.1. Strictly Functional. We use minimal cookies. These are tiny text files that exist purely to keep you logged into your account during your active session and remember your theme preferences.

15.2. No Creepy Tech. We absolutely refuse to implement third-party advertising cookies, fingerprinting scripts, or cross-site trackers. What you do outside of Heal Her is none of our business.