Verifiable private AI inference. Built to forget, not just promise to.
Why LETHE exists
Every AI request today leaves a paper trail: logged prompts, tied to an account, protected by a privacy policy you can’t verify. LETHE is built on a different premise: privacy should be structural, not promised. Read the three-pillar case, and see honestly what’s shipped versus what’s still ahead.
User Guide
Use the chat and Studio web app, guest mode, accounts, and wallet linking.
Developer / API
Call the OpenAI- and Anthropic-compatible endpoints from cURL, an SDK, or an agent framework.
API endpoints
Two dialects, one gateway. Auth, model ids, quota, and error shape are shared across all four.
Chat Completions
POST /v1/chat/completions: OpenAI-compatible, streaming supported.Responses API
POST /v1/responses: OpenAI Responses API-compatible.Messages API
POST /v1/messages: Anthropic-compatible, powers the Claude Code integration.Models
GET /v1/models: list the two public model ids.Models
Pass the id as your model parameter. Underlying provider routing is never exposed.
lethe
Text model. 1M context. Aliased from Claude Code’s
opus/sonnet/haiku picker ids automatically.lethe-vision
Multimodal model. Image and video input.
Integrations
Claude Code
Codex CLI
Vercel AI SDK
LangChain
CrewAI
Common tasks
Get an API key
Link a wallet at
/cli to raise your daily quota and get a portable ltk_ key.Make your first call
Drop-in
baseURL for OpenAI and Anthropic SDKs, plus raw cURL.Generate images
Studio: generate, edit, combine, upscale, and remove backgrounds.
What's live vs. planned
TEE-based confidential inference, x402 payments, and the $LETHE token are on the roadmap. See what’s shipped today.