LLM Proxy

Usage

Usage metadata without storing prompts or responses

Track managed-tenant usage signals while excluding prompt, transcript, audio, response, and secret content.

Teams that need AI usage visibility without making the usage store a sensitive content log.

The problem

Usage dashboards can accidentally become stores of prompts, transcripts, uploaded audio names, model responses, provider keys, or generated secrets.

How LLM Proxy helps

LLM Proxy records operational metadata for managed-tenant usage while excluding raw content and secret material.

How it works

  1. Authenticate managed proxy requests with generated tenant secrets.
  2. Record endpoint, provider, model, status, success, canonical outcome code, latency, and token counts.
  3. Exclude prompts, audio, transcripts, responses, tenant secrets, and provider keys from usage events.
  4. Expose aggregate summaries to users and admins, and safe per-failure metadata only to the owning user.

Feature-to-benefit table

Feature Why it matters here Example
Canonical outcomes Every event stores one bounded code such as invalid_request, rate_limited, request_timeout, or upstream_error. Dashboards do not persist raw error strings.
Tenant isolation Signed-in owners can inspect their own safe failure rows. Admins remain aggregate-only and foreign tenant ids return the same not-found response.
Bounded diagnostic rows Failure pages contain only occurred_at, endpoint, provider, model, status_code, outcome_code, and latency_ms. Operational investigation does not become content export.

Use-case examples

Privacy-conscious dashboard

A user sees request counts and token totals but no prompt body.

Admin overview

An admin sees tenant facts and usage summaries without generated secrets.

Historical migration

Existing status codes receive normalized outcome codes without reconstructing old provider messages.

Objections and limitations

  • Metadata can still be operationally sensitive and should be protected.
  • The dashboard is not a complete audit log.
  • Usage recording applies to managed tenants using generated-secret routing.

FAQ

What is the main job of usage metadata without prompts?

LLM Proxy records operational metadata for managed-tenant usage while excluding raw content and secret material.

Who should read this usage resource?

Teams that need AI usage visibility without making the usage store a sensitive content log.

Does this page claim provider performance or pricing advantages?

No. The supported claim is about LLM Proxy's documented routing, configuration, management, security, usage, and deployment contracts. Provider cost, speed, rankings, and benchmark claims are not made here.

Where should setup details come from?

Use the main README and implementation notes for current command, config, and endpoint details. This page summarizes the workflow without replacing LLM Proxy documentation.

What should I read next?

A closely related resource is Normalized token usage metadata across providers, which covers normalized token usage metadata.

Related resources

Use this pattern in LLM Proxy

Start from the canonical API reference, then use the management surface when the workflow needs tenant or provider configuration.

Open API reference