Skip to content

Latest commit

 

History

History
45 lines (35 loc) · 3.44 KB

File metadata and controls

45 lines (35 loc) · 3.44 KB

Charon documentation

Blind, end-to-end-encrypted LLM inference, paid in bitcoin. The relay never reads your prompts.

Documentation Description
Overview Introduction to Charon and its architecture
Quickstart Run the whole marketplace locally in dev mode
Provider guide Connect your Ollama models to the marketplace
Consumer guide Configure the OpenAI-compatible proxy and Nemesis8
Gateway deployment Deploy the gateway to Google Cloud Run
API reference Gateway HTTP endpoints and consumer endpoints
Setup checklist Action items and manual setup checklist

System architecture

Charon is a blind, paid matchmaking relay between AI consumers (like coding agents) and GPU providers. It solves the privacy and custody issues of centralized LLM providers by ensuring the matching gateway never reads the prompts or completions.

The system consists of three main components:

Gateway

The gateway (charon-gateway) is a blind WebSocket relay and control-plane server. It maintains directories of registered providers, aggregates ratings, quotes fees, verifies payments, and forwards encrypted packets between consumers and providers. Note: The gateway cannot read the contents of the prompts or completions. All payload data is end-to-end encrypted.

Provider

The provider daemon (charon provider) runs adjacent to an LLM engine (typically Ollama). It registers with the gateway, advertises supported models, and processes incoming encrypted inference requests over an outbound-only WebSocket connection. It has no open inbound ports.

Consumer

The consumer proxy (charon consumer) runs locally on the client host or as a sidecar inside a Nemesis8 container sandbox. It exposes a standard OpenAI-compatible API to local agents, handles payments, establishes end-to-end encrypted Noise sessions directly with chosen provider keys, and streams decrypted completions back to the agent.

Specification documents

For deep implementation details, see the original specification files:

  • 00 — Overview
  • 01 — Architecture
  • 02 — Identity & Auth
  • 03 — Wire Protocol
  • 04 — Encryption
  • 05 — Payments
  • 06 — Provider
  • 07 — Consumer & Nemesis8
  • 08 — Discovery & Reputation
  • 09 — Gateway
  • 10 — Security Threat Model
  • 11 — Deployment
  • 12 — UI & Dashboard