PromptCloak
LIVEIntelligence without surveillance
PromptCloak is a privacy-first AI chat app for iPhone and iPad (Mac version coming soon). Access six powerful AI models — including uncensored Grok with a 2M token context window — without ever creating an account. Your identity is a cryptographic keypair generated on your device. All conversations are stored locally, prompts are routed through a zero-knowledge stateless backend, and on-device PII scrubbing strips personal information before anything leaves your phone. Bring your own API keys for Claude and GPT to bypass our servers entirely.
Five Layers of Privacy
No Account Required
No email. No phone. No sign-up. Open the app and start chatting.
Device-Only Storage
All chat history stored locally on your device via SwiftData. Nothing is backed up to any server.
Ed25519 Crypto Signing
Every request signed with military-grade elliptic curve cryptography — the same algorithm used in SSH and Signal.
TLS 1.3 + Cert Pinning
Transport security with pinned certificates prevents man-in-the-middle attacks.
On-Device PII Scrubbing
Strict Mode strips names, places, and organizations before anything leaves your phone. You see real text. The AI sees placeholders.
Zero-Knowledge Backend
No database. Messages pass through and are immediately discarded. There is nothing to hack, subpoena, or leak.
Screenshots
Features
Zero-Knowledge Identity
No accounts, no email, no phone number. An Ed25519 keypair generated on your device is your only identity. Our servers never know who you are — and never can.
Six AI Models
Llama 3.3 70B, Llama 3.1 8B, GPT-OSS 120B, GPT-OSS 20B, Qwen 3 32B, and Grok 4 Fast (uncensored, 2M context). Switch models mid-conversation. Up to 1,000 tokens per second.
On-Device PII Scrubbing
Strict Privacy Mode uses Apple's NLTagger to detect and replace names, organizations, and locations with anonymous tokens before prompts leave your device. You see real text — the AI sees placeholders.
Bring Your Own Key
Connect your own OpenAI or Anthropic API keys for direct device-to-provider communication. BYOK requests bypass our backend entirely — zero intermediaries.
Vision & Image Analysis
Attach photos from camera, library, or drag-and-drop. All EXIF metadata is stripped by re-rendering through a graphics context — GPS, camera info, and timestamps are scrubbed before the AI sees your image.
Adaptive Memory System
Automatic memory extraction learns your preferences across conversations. Hybrid relevance ranking combines on-device NLEmbedding vectors, word overlap, and recency scoring with 50-memory cap and LRU eviction.
Custom Instructions & Skills
Persistent instructions shape every response. Create reusable Skills — custom slash commands with system prompts and template messages that sync across devices via iCloud.
Perplexity Web Search
Real-time web search powered by Perplexity Sonar. Smart auto-routing detects queries needing live data. 100 searches per month on the base plan.
Context Compression
Intelligent token budgeting with LLM-generated rolling summaries. When conversations exceed the context window, an incremental summary preserves key information while freeing space.
Meeting Transcription
Record meetings or import audio files for on-device transcription via Apple Speech. Audio never leaves your device. Transcripts appear in chat — ask the AI to summarize, extract action items, or answer questions about what was discussed.
Military-Grade Transport
TLS 1.3 minimum with SPKI certificate pinning against Let's Encrypt intermediates. Every API request is Ed25519-signed — the same cryptography used in SSH and Signal Protocol.
iPhone & iPad · Mac coming soon
Universal SwiftUI app built for iPhone and iPad. Split-view sidebar on iPad, haptic feedback on iPhone, dark mode only. The Mac version is in development and will launch shortly after.
PromptCloak vs. ChatGPT
| ChatGPT | PromptCloak | |
|---|---|---|
| Account required | Yes (email + phone) | No |
| Chat stored on server | Yes (OpenAI servers) | Never — device only |
| Data used for training | Yes (opt-out available) | Impossible (stateless) |
| PII protection | None | On-device scrubbing |
| Request auth | OAuth tokens | Ed25519 signatures |
| AI models | 1 family (GPT) | 6 models, 2 providers |
| Inference speed | ~50-80 tok/s | Up to 1,000 tok/s |
| Max context | 128K tokens | Up to 2M tokens |
| Image metadata | Retained | EXIF stripped |
| Backend data | Indefinite retention | No database exists |
Learn More
Full Feature Guide
Getting started, models, memory, slash commands, platform features, and more.
WHITE PAPERPrivacy Architecture
Technical deep dive into the five-layer privacy model, threat analysis, and honest limitations.
SETUP GUIDEClaude API Key
Step-by-step instructions to configure Anthropic Claude for direct, private AI chat.
SETUP GUIDEHermes Agent
Connect your own Hermes Agent for private, self-hosted AI with memory and tools.