Skip to the policy
Global Privacy & Safety Architecture

Pandux AI Privacy Policy & Safety Framework

Last Updated & Effective Date: November 8, 2026. Zero-custody architecture, offline local neural execution, Google Play Generative AI compliance, Pandux AI Cloud & Google Sign-In data disclosures, optional Firebase RTDB cloud sync, and user sovereign rights.

Looking for NexAI? NexAI has ended and Pandux AI is its successor. The former NexAI privacy page now redirects here.

  • On-device by defaultChats, generated media and API keys stay on your phone.
  • Works fully offlineLocal models run with zero bytes sent over the network.
  • Cloud backup is opt-inOff until you enable it, and erasable in one tap.
  • Adults only · 18+Generative tools are not offered to anyone under 18.

16 sections · search filters them as you type

1. Introduction & Overview

Privacy First

Welcome to Pandux AI ("we", "our", or "us"). Pandux AI is an advanced mobile artificial intelligence ecosystem providing conversational assistants, agentic reasoning, offline on-device neural language models, creative media studios (Image Lab, Video Studio, Voice Synthesis), Deep Research, and Team Spaces.

We operate on a transparent, Zero-Custody Architecture. Your private thoughts, personal prompts, confidential documents, and generated media belong exclusively to you. We do not sell, monetize, broker, or inspect your personal conversations, and we do not train artificial intelligence models on your private messages.

This Privacy Policy provides clear and transparent disclosures regarding how data is handled across our mobile application, our strict compliance with the Google Play Console Developer Program Policies (including the Generative AI Policy and User Data Policy), on-device neural execution, cloud inference, optional Google Sign-In and Firebase Realtime Database cloud backup, and your sovereign rights.

2. Zero-Custody Architecture & Data Partitioning

Sovereign Data

Pandux AI eliminates central telemetry and conversation harvesting by cleanly partitioning all data processing across device and cloud layers:

Local Storage (Exclusively on Device)
  • Chat histories, persona logs, and conversation threads
  • Downloaded offline GGUF local model weights
  • Persistent image and video generation media history
  • Custom BYOK API keys (encrypted in device Keystore/Keychain)
  • User UI preferences, theme choices, and daily quota trackers
Cloud Processing Layer (User Initiated Only)
  • Pandux AI Cloud Edge inference requests (authenticated via Google Sign-In)
  • Direct BYOK AI inference (OpenAI, Anthropic, Gemini, Groq, DeepSeek)
  • Optional Firebase Realtime Database cloud chat backup (strictly user opt-in)
  • Google AdMob ad serving in free tier (governed by UMP GDPR consent)
  • Anonymous system announcements via Firebase Cloud Messaging

3. Offline Local Neural LLM: 100% On-Device Processing

Zero Data Leaves Device

Pandux AI features a built-in, 100% offline local neural execution engine powered by llama.cpp and the fllama runtime:

Zero Telemetry & Offline Confidentiality
When selecting an Offline Local Model (such as SmolLM2, Llama 3.2, DeepSeek R1 Distill, Gemma 2, or Mistral), all mathematical calculations and token generation occur strictly on your physical device CPU and RAM. Zero bytes of prompts, responses, or metadata are ever transmitted over the network. You can run complete conversations in Airplane Mode with Wi-Fi and Cellular data disabled.

Hardware Optimization & Stability: On-device inference utilizes ARM NEON and SIMD vector instructions for universal compatibility and thermal protection. Model weights are verified using SHA-256 cryptographic checksums and GGUF magic-byte validation before mounting into memory to prevent unauthorized or corrupted binaries.

4. Local Network Server & Pocket Inference Node (Cloud API Bridge)

OpenAI-Compatible · Opt-In

Pandux AI includes two optional developer networking modes that let you connect IDEs, bots, and desktop apps to your phone's on-device GGUF engine. Both modes are disabled by default and only run while you explicitly toggle them on:

  • Local Network Server (LAN API): Exposes an OpenAI-compatible HTTP server on your local Wi-Fi network (default port 8080) supporting /v1/models, /v1/chat/completions, and /v1/completions. Requests require a locally generated Bearer token and never leave your LAN.
  • Pocket Inference Node (Cloud API Bridge): Opens an outbound encrypted WebSocket (WSS) tunnel from your device to the Pandux Relay using a scoped node key (pdx_node_...). Prompts and streamed tokens pass end-to-end through the tunnel to your device for inference and are never persisted or logged on relay servers.
  • Hardware Guardrails: Automatic thermal, battery, and concurrency limits protect your device; if thresholds are exceeded, the node returns standard HTTP 429 or 503 guardrail responses.

5. Universal Multi-Language Input Safety & Profanity Defense

Client-Side Moderation

To foster safe, constructive interactions and protect users from severe online toxicity, Pandux AI incorporates an active, client-side input safety filter across all chat bars, persona dialogues, and agentic workflows:

Immediate Send Blocking & Alert Notification
If abusive, severely vulgar, hateful, harassing, or sexually explicit keywords are detected in any input bar, the submit button is automatically blocked and disabled. A non-intrusive alert informs the user: "Inappropriate language detected. Please keep prompts respectful."

Multi-Language & Obfuscation Resilience: The detection engine covers 12+ international languages, including English, Spanish, French, German, Italian, Portuguese, Bengali (বাংলা script and Romanized), Hindi/Urdu (हिन्दी script and Romanized), Arabic, Russian, Chinese, and Japanese. Evaluation includes word boundary checking and leetspeak de-obfuscation (@ → a, $ → s, 0 → o, etc.) to prevent false positives.

Privacy Preserved: Profanity and toxicity evaluations are executed locally on your device before transmission. Keystroke buffers are never streamed to third-party surveillance servers.

5. Google Play Generative AI Policy & 18+ Media Safety

Strict 18+ Only

Pandux AI strictly adheres to the Google Play Console Generative AI Policy across all visual and media tools (Image Lab, Video Studio, Voice Synthesis). We maintain a zero-tolerance policy regarding the creation, distribution, or facilitation of harmful, illegal, or sexually explicit content.

Prohibited Categories Under Generative AI Guidelines
Users are strictly prohibited from generating, rendering, requesting, or distributing:
  • Adult pornography, explicit sexual acts, non-consensual sexual content, or nudity.
  • Child sexual abuse material (CSAM) or any form of child exploitation (reported immediately to law enforcement authorities including NCMEC).
  • Extreme violence, gore, self-harm, suicide encouragement, weapons creation, or acts of terrorism.
  • Non-consensual deepfakes, deceptive impersonations, harassment, defamation, or political manipulation.
Pre-Generation Content Screening

All visual generation requests pass through AiContentSafetyService with mandatory negative prompt injection (suppressing adult themes, violence, and gore) before submission. Unsafe prompts are blocked on-device.

In-App Reporting Mechanism

In compliance with Google Play Generative AI rules, every generated image, video, and text response features a one-tap reporting dialog (AiContentReportDialog). Flagged content is quarantined from history and reviewed for safety enforcement.

Removal of Unmoderated Third-Party Generators: Pandux AI has completely eliminated unmoderated third-party canvas fallback routes (including Pollinations neural canvas fallbacks). All image generation routes exclusively through verified, safety-compliant engines: Google Imagen 3 (imagen-3.0-generate-002), OpenAI DALL-E 3, and authenticated BYOK endpoints.

6. Pandux AI Cloud Intelligence & Google Authentication

Google Sign-In

To provide high-performance conversational intelligence without requiring users to configure personal API keys, Pandux AI offers Pandux AI Cloud Models powered by Groq LPUs (Llama 3.3 70B, Llama 3.1 8B Instant), OpenRouter (Llama 3.3 70B, Claude 3.5 Sonnet, DeepSeek V3), and Bynara AI (Agnes 2.0 Flash, Agnes 2.5 Pro).

Google Sign-In Authentication: Accessing Pandux AI Cloud models requires signing in with your Google account. We collect only essential profile information provided by Google Play Services:

  • Google User ID: Unique alphanumeric identifier used to associate your account with your cloud credit balance.
  • Email Address: Used for account identity, service communications, and abuse prevention.
  • Display Name & Avatar: Used solely to personalize your profile header inside the application.

Cloud Credit Ledger & Billing Transparency: Pandux AI Cloud models operate on a transparent credit balance system:

  • Upon initial sign-in, users receive a complimentary sign-up credit grant (default: 50 Credits) to explore cloud models.
  • Free cloud models (such as Llama 3.3 70B and Agnes 2.0 Flash) require 0 credits.
  • Premium frontier models (Llama 3.3 70B, Claude 3.5 Sonnet, DeepSeek V3) deduct a clearly stated credit amount per interaction.
  • All credit deductions, bonus allocations, and admin grants are recorded in a persistent transaction ledger viewable by the user.

Cloud Maintenance Transparency: When backend inference routes undergo maintenance, the app displays a clear system maintenance banner with custom administrative status messages rather than failing silently.

7. Optional Firebase Realtime Database (RTDB) Cloud Backup

Strictly Opt-In

For signed-in Google users who desire cross-device synchronization and disaster recovery, Pandux AI provides an optional chat backup feature powered by Google Firebase Realtime Database (RTDB):

Strictly Optional & Zero-Custody by Default
Cloud backup is disabled by default. Your chat sessions remain purely on your physical device unless you explicitly navigate to Settings and enable the "Auto Cloud Backup" switch or tap "Backup Now".

Cloud Storage Scope: When enabled, chat session metadata and message transcripts are synchronized to an isolated Firebase RTDB node under /user_sync/{userId}. Security rules in database.rules.json strictly enforce that only the authenticated user matching the Google User ID can read or write to their node.

Google Play User Data Deletion Compliance: In strict compliance with Google Play's User Data Deletion policy, users have full autonomous control to permanently erase their cloud backup:

  • Tapping "Delete Cloud Backup Data" in Settings immediately executes an authenticated HTTP DELETE request to permanently wipe the /user_sync/{userId} node from Firebase RTDB.
  • Deleting cloud backup data does not delete your local on-device chats unless you also select "Delete All Chats".

8. Local Media History Storage & Gallery Privacy

On-Device Storage

Pandux AI features persistent local media storage for all images and videos generated in Image Lab and Video Studio via MediaHistoryStorageService:

  • Persistent Across Restarts: Generated photos and videos are stored in your device's isolated app storage sandbox. You can exit, restart, or update the app without losing your creations.
  • Zero Unauthorized Uploads: Generated visual media is never uploaded to central gallery servers or public feeds without your explicit instruction.
  • User Takedown & Deletion: You can delete any generated image or video card from your history with a single tap, which immediately frees storage space and removes the cached asset.

9. Third-Party Cloud AI Providers & BYOK Hardware Encryption

Hardware Keystore

Pandux AI allows users to connect their own API credentials (Bring Your Own Key - BYOK) for third-party AI frontier providers, including OpenAI (GPT-4o, DALL-E 3), Anthropic (Claude 3.5 Sonnet), Google Gemini / Imagen 3, Groq, DeepSeek, OpenRouter, Cerebras, Mistral, SambaNova, Cloudflare Workers AI, and custom OpenAI-compatible endpoints.

Hardware-Backed Keystore Encryption: When you enter an API key, it is encrypted on-device using the Android Keystore / iOS Keychain (AES-256-GCM). Your keys are never transmitted to Pandux AI or stored on intermediate servers.

Direct Point-to-Point TLS 1.3: In BYOK mode, prompt and completion payloads travel directly between your device and the provider's official API servers over secure, authenticated TLS 1.3 channels. Data handling by those providers is governed by their respective developer privacy policies.

10. Advertising, Cloud Messaging & Permissions Breakdown

Transparent Disclosures

Google AdMob (Free Tier Only): To support free access to basic tools, Pandux AI displays advertisements served by Google AdMob in the free cloud mode:

  • AdMob may process device identifiers (such as Android Advertising ID / AAID), app diagnostic metrics, and coarse location to serve ads and prevent fraud.
  • Users in the European Economic Area (EEA) and the UK receive the Google User Messaging Platform (UMP) consent form, allowing granular opt-in/opt-out of personalized advertising in compliance with GDPR.
  • BYOK and Local LLM modes are 100% ad-free; ad initialization is completely suppressed when your own keys or local models are active.
  • Pandux AI never shares prompt contents, transcripts, or user media with advertising networks.

Firebase Cloud Messaging (FCM): Used exclusively to deliver system maintenance notices, model weight releases, and critical security broadcasts. Broadcasts use anonymous topic subscriptions (e.g. /topics/announcements) without user profiling.

Device Permissions Rationale: Pandux AI requests only permissions strictly necessary for user-initiated actions:

  • POST_NOTIFICATIONS (Android 13+): Opt-in permission to deliver offline model download completion updates and system maintenance notices.
  • RECORD_AUDIO: Accessed solely during the exact duration you tap/hold the voice chat microphone. Audio is processed on-device or streamed to your selected provider; it is never recorded in the background.
  • Storage & Media Access: Scoped access required only when you explicitly select a file/document for AI analysis or save an image/video to your device gallery.
  • INTERNET & ACCESS_NETWORK_STATE: Required to establish encrypted connections for cloud inference, Google Sign-In, and optional Firebase sync.

11. Google Play Data Safety Declarations Table

Console Aligned

The following table mirrors our formal disclosures in the Google Play Console Data Safety section:

Data CategorySpecific Data TypesCollected?Shared?PurposeEncrypted in TransitUser Deletion Option
Personal InfoGoogle User ID, Name, Email (only if signed in with Google)Yes (Optional)NoAccount authentication, Pandux AI Cloud credit ledger, abuse preventionYes (TLS 1.3)Yes (Direct In-App)
Messages & ChatsUser prompts, AI responses, chat titlesOptionalNoStored locally by default; synced to Firebase RTDB node /user_sync only if user turns on Cloud BackupYes (TLS 1.3)Yes (Instant Wipe)
Photos & VideosGenerated AI images, videos, storyboardsNo (Local Only)NoStored exclusively in private on-device sandbox storage (MediaHistoryStorageService)N/A (On-Device)Yes (Single/All Delete)
Audio FilesVoice recordings for speech-to-textEphemeralNoStreamed in real-time during microphone button press; discarded immediately after transcriptionYes (TLS 1.3)Auto-Discarded
Files & DocsImported PDFs, images, text for analysisNo (Local Only)NoProcessed locally on device or streamed ephemerally to selected BYOK providerYes (TLS 1.3)Auto-Cleared
Device / Ad IDsAndroid Advertising ID (AAID), IP addressYes (Free Mode)Shared (AdMob)Ad serving, fraud prevention, and frequency capping (Suppressed in BYOK/Local mode)Yes (HTTPS)Yes (Device Settings)
Crash & DiagnosticsAnonymous crash stack traces, performance metricsYes (Optional)NoApp stability diagnostics and bug resolution (no message contents attached)Yes (TLS 1.3)N/A (Anonymized)

12. Age Policy & Children's Online Privacy (COPPA & GDPR-K)

Adults Only

Pandux AI is strictly intended for individuals aged 18 years or older. Due to the sophisticated capabilities of generative multimodal artificial intelligence, we do not market our application to, nor knowingly collect personal data from, children under the age of 18.

In accordance with the U.S. Children's Online Privacy Protection Act (COPPA), the EU General Data Protection Regulation (GDPR-K), and Google Play Families Policies, if a parent or legal guardian becomes aware that a minor has provided data or accessed generative tools, please contact us at [email protected] for immediate data purge and account revocation.

13. Your Sovereign Rights & Erasure Controls (GDPR / CCPA / LGPD)

Universal Control

Regardless of your geographic location, Pandux AI provides universal data sovereignty tools honoring GDPR, CCPA, and Google Play policies:

Right to Instant Local Erasure (Delete All Chats)
You can permanently wipe all local conversation archives, generated media history, and cached documents by tapping "Delete All Chats" in App Settings or clearing application data in Android System Settings.
Right to Erase Cloud Backup Data (Google Play Compliant)
If you enabled Firebase RTDB cloud sync, tapping "Delete Cloud Backup Data" in Settings permanently deletes your node from /user_sync/{userId} on Firebase RTDB via an immediate authenticated HTTP DELETE request.
Right to Export & Data Portability
You can copy, share, or export any chat thread, Deep Research report, or transcript to your clipboard, plain text, Markdown, or PDF format at any time.
Right to Purge Stored API Keys & Offline Weights
You can delete any saved BYOK key from device Keystore/Keychain with a single tap in Provider Settings. Downloaded GGUF neural models can be removed from disk in the Local Model setup screen to instantly reclaim storage.
Request account & data deletion without the app
If you no longer have Pandux AI installed, email [email protected] from the Google account address you signed in with and ask us to delete your Pandux AI account and cloud backup data. Chats, media and API keys that were only ever stored on your device are removed when you uninstall the app or clear its data.

15. Developer & API Quick Reference

OpenAI SDK Compatible

Developers can connect Cursor, Continue.dev, LangChain, OpenAI SDKs, or custom scripts directly to Pandux AI via the Local Network Server or Pocket Inference Node.

For complete endpoint specifications, curl/Node.js/Python quickstarts, hardware guardrail error codes, and prompt hygiene guidelines, see the dedicated Pandux AI Developer & API Reference.

16. Contact Us & Safety Officer

Always Available

If you have any questions, suggestions, or concerns regarding this Privacy Policy, our multi-language profanity safety filter, offline local models, Pandux AI Cloud, or generative media safety, please reach out to our dedicated engineering and privacy team:

Data Privacy Officer
[email protected]

Inquiries on data sovereignty, encryption, Firebase RTDB sync, and GDPR/CCPA rights.

Trust & Content Safety
[email protected]

Inquiries on prompt moderation, 18+ policy, content takedowns, or user safety reports.

Developer Organization: Bengal Bytes / Pandux AI Core Team