Local AI privacy. Keep sensitive data on your machine.
A local privacy gateway that scrubs sensitive data before prompts leave your machine.
Aelvyril detects secrets in copied prompts, replaces them with safe tokens for transit, then restores real values in AI responses automatically.
The Cycle
Watch the full protection flow: identify sensitive text, mask it before transit, and restore true values only after the model responds.
Scanning your prompt...
Every prompt can carry secrets.
Three Pillars
Aelvyril protects the full request-response loop: sensitive text is detected, masked before transit, then restored only on your machine.
Monitors clipboard and browser inputs for sensitive patterns in real time.
Replaces PII with context-aware tokens before any data leaves your machine.
Rehydrates the LLM response with real values, seamlessly and automatically.
Watches for secrets, API keys, and credentials in your clipboard.
Context-aware tokenization via NER for names, locations, orgs.
Works with any LLM — OpenAI, Anthropic, Gemini, local models.
OS keychain storage. No keys in config, logs, or dumps.
Runs on your machine. No external deps.
Nothing logged. Token maps live in memory only.
Works offline. No server transit.
Audit the code. Run your own build.