One key, every major model, and nobody upstream knows it is you. Swap a base URL and your prompts stop being an account record.
base_url = "https://api.shadr.io/v1"Swap one line.
Point the OpenAI or Anthropic SDK at us and keep everything else. Claude, GPT, Gemini, DeepSeek and the open models behind one key and one bill. Claude Code, Cursor and Aider work unchanged.
Keys that can't overspend.
Mint a key per person, per agent, per CI job. Give each one a monthly ceiling and the models it may touch. When the ceiling hits, that key stops. Nothing else does.
Nothing to subpoena.
Our gateway stores account and billing records, not prompts or replies. Requests use our credentials, and supported text calls request secret masking. Sign in with Privy and pay in USDC on Base through your wallet.
How a request travels.
Same JSON, different host
The request you send today, to api.shadr.io. Your key tells us whom to bill and nothing more.
Identity stops here
Client headers, IP and account are dropped at the edge. API keys, tokens, seed phrases and passwords inside the prompt are swapped for placeholders.
One of thousands
The model provider sees a request from Shadr, indistinguishable from the rest of that minute's traffic. The reply comes back, placeholders are restored, and we write down a token count.
Every model that matters.
Claude Opus 4.8 · GPT-5.5 · Gemini 3.1 Pro
Reasoning, tools and vision, routed to a healthy provider every time.
Claude Haiku 4.5 · GPT-4o mini · MiniMax M2.1
For the background calls that make up most of an agent's bill.
DeepSeek V4 Pro · GLM 4.7 · GPT-OSS 120B
Cheap, capable, and served by providers that publish what they keep.