The Models
Every leading model, none of the exposure.
One private endpoint, routed to the best models from every major lab — text, code, image, and more — all behind the same stateless, unlogged gateway.
What Gets Kept
Same question. Very different paper trail.
Typical:An account, an email, a card on file.
LETHE:Nothing. Ask right away, or connect a wallet for more.
Typical:An API key that points straight to you.
LETHE:A single-use token that points to no one.
Typical:Logged and kept, often for 30 days or more.
LETHE:Passed to the model, then dropped. Never stored.
Typical:On their servers, tied to your account.
LETHE:Only in your browser, and only if you save it.
Typical:A monthly bill with your name on every call.
LETHE:USDC for a batch of tokens. The payment never meets a prompt.
Typical:A privacy policy.
LETHE:Tokens you can check yourself, in your browser.
The Protocol
How a request runs inside LETHE, tied to no one.
Unlinkable request
Pay in USDC via x402 on Base or Solana, or use your wallet's daily allowance, and get blind-signed tokens. Each token pays for one call and can't be traced back to the wallet or payment that bought it.
Attested executionRoadmap
Next, the model moves into a Trusted Execution Environment so even LETHE's own infrastructure can't read the prompt. Today, requests pass through a stateless proxy and are never stored.
Verifiable deletionRoadmap
Planned: an enclave attestation proving each request was discarded. Today, no conversation is kept server-side and guest history never leaves your browser.
See It Run
One request in, one response out, nothing remembered.
One OpenAI-compatible endpoint, callable from cURL, TypeScript, or Python. Stateless and unlogged. Pay each call with a single-use private token, or point your SDK at lethe-proxy and it spends tokens for you.
Connect for CLI →// lethe-proxy spends a private token per call
const client = new OpenAI({
baseURL: "http://127.0.0.1:8402/v1",
apiKey: "local",
});
const res = await client.chat.completions.create({
model: "lethe",
messages: [{ role: "user", content: prompt }],
});
For Agents
Give your agent a line that isn't tapped to you.
Agents send more context than people do: code, files, credentials by accident. Route them through LETHE and every call arrives with no key, no wallet and no account behind it. Fund them with a wallet's daily allowance, or let them buy their own tokens in USDC over x402.
Add LETHE to any MCP client
Claude Code, Cursor or Claude Desktop get a lethe_ask tool for anything sensitive. Each call spends one private token.
claude mcp add lethe \ -e LETHE_API_KEY=ltk_… \ -- node scripts/lethe-proxy.mjs --mcp
Run your whole agent through it
lethe-proxy swaps your tool's key for a fresh token on every call and tops up in the background, so purchases don't line up with requests.
npm run lethe:proxy export ANTHROPIC_BASE_URL=http://127.0.0.1:8402 export OPENAI_BASE_URL=http://127.0.0.1:8402/v1
The Token
Stake $LETHE, earn your share of every call.
Stakers earn protocol revenue. Every call burns supply. The more LETHE is used, the scarcer it gets.
A share of every call
Stake $LETHE to earn a share of the USDC paid through x402 for every inference call on the network.
Deflationary by design
A portion of $LETHE is burned with every call, permanently shrinking supply as usage grows.
Our Product
Private inference infrastructure for agents and developers.
Cognition
Your prompts and completions are never stored — and a private token carries nothing that points back to you.
Perception
No dashboard, no request logs, no stored conversations — not even for us.
Timing
Tokens are bought in bulk via x402 and spent one call at a time — the payment and the prompt never meet.
Network
Built for agents talking to agents, machine to machine, over the OpenAI API.
Begin
Your first request can already be private.
No account to create, no waitlist to join. Buy private tokens in USDC via x402 today — requests that carry no identity at all.