TokenHub Server

A Python aiohttp WebSocket server that coordinates token exchanges between LLM API providers.

Overview

The server matches clients offering and requesting specific model access. It calculates exchange rates based on real-world input and output pricing for 20 models from OpenAI, Anthropic, and Gemini.

Features

• Peer matching based on model availability and demand.
• Real-time exchange rate calculation from pricing.py.
• Usage report relaying between connected peers.
• Support for advanced token budgeting (separate input/output rates).

Requirements

• Python 3.11+
• aiohttp >= 3.11.0

Installation

pip install -e .

Test: python -m pytest -q

Running the Server

Start the server using the entry point:

tokenhub-server

Or run the module directly:

python -m server.main

Docker

Build and run via Docker:

docker build -t tokenhub-server .
docker run -p 8080:8080 tokenhub-server

Configuration

Configure the server using environment variables:

• HOST: Bind address (default: 0.0.0.0)
• PORT: Listen port (default: 8080)

How the Exchange Works

The server facilitates "fair swap" matching. If Alice offers OpenAI and wants Anthropic, and Bob offers Anthropic and wants OpenAI, the server pairs them. It calculates a metered budget based on the relative cost of the models. For simple budgets, it uses total token counts. For advanced budgets, it uses specific input and output pricing.