← Back to Login

Privacy Policy

This Privacy Policy describes how Metayage Private Limited ("Company", "we", "us", or "our") collects, uses, stores, and protects your personal data when you use the IDPD — Invention Disclosure Platform ("Software"). By accessing or using the Software, you agree to the practices described in this policy.

1. Data We Collect

1.1 Account & Identity Data

1.2 Invention Disclosure Data

1.3 Usage & Technical Data

2. AI Features & How They Work

2.1 AI is Optional — You are in Control

All AI-assisted features are entirely optional. If you do not wish to use AI generation, simply leave AI disabled or turn it off at any time via Settings → AI Preferences. The platform's core disclosure workflow — submitting inventions, tracking status, collaborating — continues to work without AI enabled. If AI is turned off, no invention data is sent to any external AI provider.

How to opt out: Go to Settings → AI Preferences and turn off "AI Provider Features". Your preference is saved instantly. You may re-enable AI at any time.

2.2 How AI Generation Works (Data Flow)

When you request an AI-assisted output (e.g. patent claims draft, invention summary), the following happens:

  1. Optional ML enrichment (local, on-server): If Local ML Enrichment is enabled, your invention text is first processed by our on-premise ML models — BAAI/bge-base-en-v1.5 (BGE), KeyBERT, and YAKE — running on our DigitalOcean VPS. This step extracts keywords and finds similar inventions from your organisation's historical disclosures. This data never leaves the server.
  2. Prompt construction: A structured prompt is built from your invention content (and optionally the ML-enriched context). Sensitive identifiers may be included only as needed for the task (e.g. invention title).
  3. API call to external AI provider: The prompt is sent via HTTPS to one of the configured AI providers (see Section 5). The provider generates a response which is streamed back to your browser in real time.
  4. Display only: The generated output is displayed to you. It is your choice whether to save or discard it. Nothing is auto-saved on your behalf.

2.3 AI Models Used

Depending on the generation mode selected, the following models may be used:

Mode Provider Primary Model Fallback Model
Fast Groq llama-3.3-70b-versatile llama-3.1-8b-instant
Balanced Google Gemini gemma-4-31b-it gemma-4-26b-a4b-it
Enhanced NVIDIA NIM mistralai/mistral-large-3-675b-instruct meta/llama-4-maverick-17b-128e-instruct
Ultra Gemini + Groq (combined) Both primary models run in parallel; responses are merged
Local (ML enrichment) On-premise (our VPS) BAAI/bge-base-en-v1.5, KeyBERT, YAKE N/A — always runs locally

3. Data Security

4. Infrastructure & Hosting

IDPD is hosted on a DigitalOcean Virtual Private Server (VPS) located in Bangalore, India. All invention data, database records, and ML model files reside on this server. DigitalOcean does not have access to your application data.

The on-premise ML pipeline (BGE, KeyBERT, YAKE, FAISS similarity search) runs entirely on this same VPS. No data is sent to an external service for local ML processing.

DigitalOcean privacy policy: digitalocean.com/legal/privacy-policy

5. Third-Party AI API Providers

When AI features are enabled and used, your invention disclosure content is transmitted to the selected AI provider's API over HTTPS. Each provider below acts as a data processor and is contractually obliged to process your data only for the purpose of returning the AI response.

Important: If you disable AI Provider Features in Settings → AI Preferences, no data is sent to any of the providers below.

Google Gemini (Gemma 4 models)

Operator: Google LLC, 1600 Amphitheatre Parkway, Mountain View, CA 94043, USA

Models used: gemma-4-31b-it gemma-4-26b-a4b-it

Data sent: Your invention disclosure text and questionnaire responses, structured as a prompt.

Data retention & training: Google's API terms state that prompts and responses are not used to train Google's models without explicit opt-in. Operational logs may be retained for up to 30 days for abuse monitoring.

Privacy policy: policies.google.com/privacy

Groq (Llama 3 models)

Operator: Groq, Inc., 101 University Ave, Suite 1400, Palo Alto, CA 94301, USA

Models used: llama-3.3-70b-versatile llama-3.1-8b-instant

Data sent: Invention disclosure content structured as a prompt.

Data retention & training: Groq does not use API request data to train or fine-tune models. API data may be retained in operational logs for up to 30 days, then deleted.

Privacy policy: groq.com/privacy-policy

NVIDIA NIM (Mistral Large / Llama 4 models)

Operator: NVIDIA Corporation, 2788 San Tomas Expressway, Santa Clara, CA 95051, USA

Models used: mistralai/mistral-large-3-675b-instruct meta/llama-4-maverick-17b-128e-instruct

Data sent: Invention disclosure content structured as a prompt.

Data retention & training: API input data is not used to train NVIDIA's models. Request metadata may be retained for up to 30 days.

Privacy policy: nvidia.com/en-us/about-nvidia/privacy-policy

Brevo — Transactional Email Delivery

Operator: Sendinblue SAS, 55 rue d'Amsterdam, 75008 Paris, France

Data sent: Recipient email address, sender name, email subject and body (notification content only).

Privacy policy: brevo.com/legal/privacypolicy

6. Data Retention

Data Type Retained By Retention Period
Account & invention data IDPD (our VPS) Until account deletion; active patent prosecution files retained for the duration of the engagement
Session data IDPD (our VPS) 2 hours idle / browser close
AI prompt & response (Google Gemini) Google LLC Operational logs up to 30 days
AI prompt & response (Groq) Groq, Inc. Operational logs up to 30 days, then deleted
AI prompt & response (NVIDIA NIM) NVIDIA Corporation Processed transiently; metadata up to 30 days
Email delivery logs Brevo (Sendinblue SAS) Per Brevo's standard policy

Upon account deletion request, we will delete or anonymise your personal data and invention disclosures within 30 days, except where retention is required for active patent prosecution or legal compliance.

7. Local ML Model Training

IDPD runs its own machine learning models on our private VPS: BAAI/bge-base-en-v1.5 (semantic embeddings), KeyBERT (keyphrase extraction), and YAKE (statistical keyword extraction). These models currently serve only to enrich AI prompts — they do not generate final output.

In the future, we may wish to fine-tune these local models using patterns from anonymised invention disclosures to improve their domain-specific performance for patent-related language. We will only use your data for this purpose if you explicitly give consent.

How to give or withdraw consent: Go to Settings → AI Preferences → ML Model Training Consent and toggle the switch. You can change your choice at any time.

8. Your Rights

You have the right to:

Contact us to exercise any right: ip@myipstrategy.com. We will respond within 30 days.

9. Cookies & Session Storage

The Software uses a single session cookie to maintain your logged-in state. It is:

We do not use advertising or tracking cookies. The Cloudflare Turnstile widget on the login page may set a functional cookie to verify human interaction; this is a security control only.

10. Changes to This Policy

We may update this Privacy Policy from time to time. Material changes will be communicated by updating the "Last Updated" date below and, where appropriate, by email to active users.

11. Contact Us

Metayage Private Limited
Email: ip@myipstrategy.com

Last Updated: June 02, 2026