poolswitch 0.1.3

Production-grade PoolSwitch proxy with quota failover and multi-language SDKs.

PoolSwitch

poolswitch is a production-oriented API key rotation toolkit that can run either:

• directly inside your app as an embedded client
• as a local or shared proxy server

It is designed for provider APIs such as OpenAI, Anthropic, Groq, Google, HuggingFace, or any generic HTTP endpoint that authenticates requests with an API key header.

Features

• Embedded Python client with built-in key rotation, retry, cooldown, and quota failover
• Embedded Node.js client with built-in key rotation, retry, cooldown, and quota failover
• Async local HTTP proxy with low overhead
• Key rotation strategies: round_robin, least_used, random, quota_failover
• Quota-aware cooldowns and automatic failover
• Retry system with exponential backoff and network/429 handling
• Pluggable storage: memory, Redis, SQLite
• YAML, environment variable, and CLI configuration
• Prometheus-compatible /metrics endpoint
• CLI commands for startup, status, metrics, and key management
• Proxy SDKs for Python, Node.js, and Go

Quick Start

Embedded Client (Recommended)

Node.js / TypeScript:

import { PoolSwitchClient } from "poolswitch-node";

const client = new PoolSwitchClient({
  upstreamBaseUrl: "https://api.openai.com",
  keys: [
    { id: "primary", value: process.env.OPENAI_KEY_1! },
    { id: "backup", value: process.env.OPENAI_KEY_2! }
  ],
  strategy: "quota_failover"
});

const response = await client.post("/v1/chat/completions", {
  json: {
    model: "gpt-4o-mini",
    messages: [{ role: "user", content: "hello" }]
  }
});

Python:

from poolswitch import PoolSwitchClient

client = PoolSwitchClient(
    upstream_base_url="https://api.openai.com",
    keys=[
        {"id": "primary", "value": "sk-123"},
        {"id": "backup", "value": "sk-456"},
    ],
    strategy="quota_failover",
    retry_attempts=3,
    cooldown_seconds=3600,
)

response = client.post(
    "/v1/chat/completions",
    json={
        "model": "gpt-4o-mini",
        "messages": [{"role": "user", "content": "hello"}],
    },
)

For async apps:

from poolswitch import AsyncPoolSwitchClient

async with AsyncPoolSwitchClient(
    upstream_base_url="https://api.openai.com",
    keys=["sk-123", "sk-456"],
) as client:
    response = await client.get("/v1/models")

Proxy Mode

pip install -e .
poolswitch start --config poolswitch.example.yaml

Example request:

curl -X POST http://127.0.0.1:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o-mini","messages":[{"role":"user","content":"hello"}]}'

See docs/architecture.md for the design and the SDK folders for client usage.

Docs Site

This repo now includes a VitePress docs site in docs.

Preview locally with:

npm install
npm run docs:dev

Build static files with:

npm run docs:build

The deployable output is written to docs/.vitepress/dist, which can be uploaded directly to Netlify, Vercel, or any static host.