Skip to main content
The proxy is designed for HTTP API keys such as OpenAI, Stripe, Resend, and Anthropic: keys sent through an Authorization or x-api-key header to an HTTPS API. The proxy can intercept and replace these values. For non-HTTP secrets such as database credentials, configuration, or DATABASE_URL, use secret(). The proxy cannot intercept a raw TCP connection.

​
How it works

In proxy mode, your application uses a {{SECRET_NAME}} placeholder for an HTTP API key. The local proxy intercepts each outgoing request, replaces the placeholder with the real value, and forwards the request upstream. 
App → http://127.0.0.1:PORT → [proxy replaces {{OPENAI_API_KEY}}] → https://api.openai.com
The real key lives inside the separate proxy process and never enters your application’s memory.

​
Request processing

1

Authentication

The proxy verifies x-elding-token using a constant-time comparison. Only the child process started by elding proxy receives this token.
2

Anti-SSRF

The target hostname is resolved. Private, loopback, link-local, and cloud metadata IP addresses such as 169.254.169.254 are blocked. DNS pinning prevents rebinding attacks.
3

Host locking

For every placeholder used, the proxy verifies that the key’s authorized domain matches the target. A mismatch returns HTTP 403 and prevents exfiltration.
4

Injection and forwarding

Placeholders are replaced in request headers, then the HTTPS request is streamed to the upstream service.

​
Proxy security

  • Binds to 127.0.0.1 only and is never exposed to the network.
  • Uses a random session token with constant-time comparison.
  • Allows HTTPS only and strips hop-by-hop headers.
  • Logs placeholder names only, never secret values.
A compromised dependency or AI agent inside your application only sees {{OPENAI_API_KEY}}, never the real key.