Your keys are securely encrypted in your browser's IndexedDB using AES-GCM. Exporting downloads the encrypted blob (you still need the passphrase to use it).
TrustShell is an open-source router that prioritizes free-tier LLM inferences (Groq, Cerebras, Gemini Flash) and falls back to paid APIs (OpenAI, Anthropic) using your own keys. Your paid keys never leave your browser unencrypted—they are sent in memory per-request and redacted from server logs.
Every prompt and answer is scored by the HAL Benchmark, updating your Agent's RepID score to objectively measure trustworthiness over time.
View source on GitHub