Ghostway
Open console

Tor-routed AI gateway

One endpoint.
Every model.
Nothing stored.

OpenAI-compatible gateway with Tor and TLS routing, per-key billing, and hedged requests that rescue slow tails.

Open fleet console Copy cURL

OpenAI

compatible API

Tor + TLS

dual routing

70 B

per log row

POST /v1/chat/completions (SSE) DIRECT TLS

$ curl gateway/v1/chat/completions \

-d '{"model":"muse-spark-1.3"}' --no-buffer

TTFB

live

TTFT

live

tok/s

live

No prompt storage

Prompts and completions never touch disk. Each telemetry row holds about 70 bytes of metadata: model, tokens, latency, cost, route. Ten million requests fit in roughly 700 MB.

id, model, prompt_tok, compl_tok, cost, latency_ms, created_at

Tor circuit pool

Pre-warmed circuits across four exit countries. Exits that return 429 or 503 are quarantined automatically.

DE Frankfurt
NL Amsterdam
US Dallas
FR Paris

Key vault and billing

Provision gw-… keys with credit balances and one-click top-ups. Masked at rest, shown once at creation.

Model comparison

Same prompt on Direct and Tor, with live TTFB, TTFT, and tokens per second. Tuning applies with no restart.


      

    

    

      

        
Drop-in OpenAI compatible

        
Point your existing client at one base URL. Keep your SDK, swap the key.

        

          
baseURL https://ghostway.pages.dev/v1

          
auth Bearer gw-…

          
stream SSE, true

        

        Provision a key
      

    

  


  
  

    

      

      
Deploy in minutes

      
One gateway
for every model.

      

      
Point your client at one URL. No content stored, ever.

      

        Open fleet console
        See the code