Theme
SUB&SUB Unified API
All systems operational · 99.97% uptime

One key, every AI subscription you have

SUB&SUB merges your ChatGPT, Claude, Bedrock, and other model subscriptions behind a single OpenAI-compatible endpoint. Change one base_url and your existing SDK keeps working — streaming responses included.

Get started View architecture
P50 latency
128ms
P99 latency
214ms
30-day uptime
99.97%
Aggregated models
42 models

Architecture

Transparent routing, end to end

Client → edge CDN → origin → sub2api → account pool → upstream model. Streaming responses travel the same path back. No vendor lock-in, no proprietary protocol.

Step 01 Client SDK · CLI · Web
Step 02 Edge CDN WAF · TLS · Rate-limit
Step 03 Origin Anycast · low-RTT
Step 04 Nginx SSE · keep-alive
Step 05 sub2api :8080 · protocol shim
Step 06 Account pool Round-robin · failover
Step 07 Upstream ChatGPT · Claude · ...
Standard hop Core component → Request path; streaming response returns the same way

Features

Tuned for streaming, end to end

SSE keep-alive, proxy_buffering off, edge smart-routing — no "stall then dump" behaviour. Every layer behaves like the upstream API would.

Anycast edge + smart routing

Edge CDN with WAF, TLS termination, and intelligent path selection. SSE-friendly: no minification, no rocket-loader, no buffering.

Global edge

Origin tuned for SSE

Nginx with proxy_buffering off, long timeouts, and HTTP/2. No surprise gateway timeouts on long-running completions.

SSE keep-alive

sub2api protocol shim

Converts ChatGPT / Claude web sessions into OpenAI-compatible API. Account pool mounts at /data, auto round-robin.

OpenAI-compatible

Real-time observability

CPU, memory, traffic, and failure rate are all charted. Per-key quota, per-model usage, account-pool health at a glance.

Prometheus + Grafana

WAF + rate-limit

Per-IP throttling at the edge: 30 req / 10 s by default. Block obvious scrapers (python-requests, curl-default UA) on sight.

Edge rules

One key, every model

A single API key fans out to GPT-4o, Claude 4.7, Bedrock-proxied models. Model name passes through — no client changes.

42 models

Upstream providers

What we aggregate today

ChatGPT Plus
Web → API · 24 accounts
GPT-4oo1
Claude Pro
Cookie relay · warming
4.74.6
AWS Bedrock
Enterprise · 0.72× rate
SonnetOpus
Kiro relay
Beta · invite-only
Code

Quickstart

30 seconds to integrate

OpenAI / Anthropic-compatible. Change one base_url and your existing SDK keeps working.


    

    

      
Note: Nginx must set proxy_buffering off, otherwise streaming responses get buffered into a single late dump. At the CDN, disable Auto-Minify, Rocket Loader, and Polish — they all break SSE.

    

  


  

    

      
Pricing

      
Pay-as-you-go, transparent

      
Top up once. No monthly fee. Balance never expires. Charged at upstream list price × your tier rate.

    

    

      

        
Starter

        
$5min top-up

        
    
          
  • OpenAI / Claude web pool
    • 
          
  • 1.0× rate · all models
    • 
          
  • 3 concurrent · SSE streaming
    • 
          
  • Community support · Discord
    • 
        

        Free trial
      

      

        
Pro

        
$50top-up

        
    
          
  • Everything in Starter
    • 
          
  • 0.85× rate · priority pool
    • 
          
  • 20 concurrent · sub-keys
    • 
          
  • Bedrock proxy · 0.72× rate
    • 
          
  • Usage dashboard · alerts
    • 
        

        Get Pro
      

      

        
Enterprise

        
Customcontact sales

        
    
          
  • Dedicated origin · isolated pool
    • 
          
  • SLA 99.95% · 24/7 response
    • 
          
  • Unlimited concurrency
    • 
          
  • Invoiced billing · MSA
    • 
        

        Contact