Skip to main content
LETHE is a chat and image-generation product in front of a developer API. Here’s how a request actually flows today.

Guest by default

You don’t need an account to use LETHE. Every browser gets a small free allowance (currently 3 messages per network) tracked without a login. Chat and Studio history for guests is stored locally in your browser (localStorage). It never leaves your device unless you sign in.

Optional account

Signing in (email magic link, Google, Discord, or an Ethereum wallet via Sign-In with Ethereum) moves your chat and Studio history into a Supabase-backed database, scoped to your account with row-level security. This gets you cross-device history, at the cost of that history living server-side rather than only on your device.
Signed-in history is stored as regular (unencrypted) rows in Postgres today. See Privacy Model & TEE for the gap between this and LETHE’s longer-term privacy goals.

Wallet linking for quota

Separately from account sign-in, you can link a wallet at /cli to raise your daily limit (50 messages/day per UTC day) and receive a portable API key (ltk_…) you can use from any machine or CLI tool. Linking is free. Paying is optional and only used to buy private tokens (below).

Private tokens

Requests can be paid with a single-use private token instead of a key. Tokens are signed blind (Privacy Pass, RFC 9578), so LETHE can check that a token is genuine but can’t link it to the wallet, key or payment that obtained it. You get tokens by buying them with USDC via x402 (Base or Solana) or by converting a wallet’s daily allowance. Signed-in web chat does this automatically and sends token-paid turns without your wallet cookie. See Private Tokens.

Model backend

Requests are routed to an upstream model provider through LETHE’s inference gateway. The gateway normalizes everything to two public model ids, lethe (text) and lethe-vision (multimodal), and strips any upstream provider names out of responses before they reach you.

What this is not, yet

The longer-term vision is inference running inside a Trusted Execution Environment (TEE), with cryptographic attestation that the request was deleted. This isn’t implemented yet. Prompts are processed in plaintext by LETHE’s gateway and the upstream model provider, and they’re never stored. See Privacy Model & TEE.