Skip to main content

​
Log the Request ID

Every response carries X-Request-ID: the one you sent (up to 128 characters, sanitized) or a fresh one (16 random bytes, hex). It correlates a user report, your application log, and a gateway-side lookup without storing content. Set your own on the way in so the IDs match end to end.

​
Metadata, Not Content

Most production issues are debuggable from:
  • The request ID.
  • The model ID requested.
  • HTTP status and finish_reason (stop, length, tool_calls, error).
  • Token counts from usage, latency, whether streaming was used.
  • The error code and message from the envelope.
Content logging belongs behind a deliberate, access-controlled, time-bounded process, never on by default.

​
Never Log the Credential

Do not log the Authorization header or the key. Many HTTP libraries dump full headers on error; redact before anything is written. Allowlist the fields you log; a blocklist fails open. See Secure API Key Handling.

​
Match the Gateway

Nito stores nothing you send or receive: prompts, completions, reasoning, tool payloads, and attachment content are not persisted, and inference responses carry Cache-Control: no-store. Logs that capture full content become the one place the data sits at rest. See Data Retention and Visibility.

​
Checklist

  • X-Request-ID on every call, set on the way in.
  • Metadata, not bodies.
  • Authorization never logged; allowlist redaction.
  • Content logging as a controlled exception.

Request IDs

The X-Request-ID header.

Privacy Overview

The four levels.

Data Retention and Visibility

What is kept, who sees what.

Secure API Key Handling

Keys.