Skip to content
TrustedRails

API Keys

API keys route your requests through TrustedRails's proxy layer, which forwards them to the inference network and returns responses in the same API format your app speaks: OpenAI or Anthropic.

How it works

Section titled “How it works”
Your app → Proxy → GPU Network → Response
  1. Your application sends a request using the OpenAI or Anthropic API format
  2. TrustedRails receives it and forwards it to the GPU network
  3. The inference runs on available GPUs
  4. The response is returned to your app in the same standard format

From your application's perspective, this is identical to calling the OpenAI or Anthropic API.

What's compatible

Section titled “What's compatible”

API keys support the OpenAI Chat Completions API, including:

  • Chat completions (/v1/chat/completions)
  • Streaming responses (stream: true)
  • System, user, and assistant messages
  • Temperature, top_p, max_tokens, and other standard parameters
  • Tool calling (tools, tool_choice)

The same keys also work over the Anthropic Messages API (/v1/messages), with Anthropic SDKs and Anthropic-native tools like Claude Code.

See OpenAI API Compatibility and Anthropic API Compatibility for the full list of supported endpoints and parameters.

Base URL

Section titled “Base URL”

For OpenAI clients, all requests go to:

https://proxy.trustedrails.com/v1

Set this as the base_url (Python) or baseURL (JavaScript) in your OpenAI client configuration.

Anthropic clients use the bare host (https://proxy.trustedrails.com, without /v1) as the base URL (see Anthropic API Compatibility).

Use cases

Section titled “Use cases”

| Use case | Works? | |---|---| | Replacing OpenAI in existing code | Yes | | Replacing Anthropic in existing code | Yes | | Using LangChain, LlamaIndex, or similar frameworks | Yes | | Using Claude Code, Cursor, or other AI-powered dev tools | Yes | | Any tool that supports an OpenAI- or Anthropic-compatible API | Yes |

Security

Section titled “Security”
  • Your key secret starts with tr-prx- and should be treated like any API credential
  • TrustedRails does not log or store your prompts and completions
  • If a key is compromised, rotate it immediately: this invalidates the old secret and issues a new one