Standardize request/app metadata across Foundry and Railway deployments

[anand-testcompare]

Background

We want consistent metadata (app id, app URL/origin, deployment environment, etc.) to be attached to LLM requests regardless of where the API is running (Railway vs Foundry Compute Modules).

Today, OpenRouter/LiteLLM usage logs can show an origin like https://litellm.ai/ and app identifiers that do not reflect our deployment. We already support OpenRouter headers via env vars (e.g. OPENROUTER_HTTP_REFERER, OPENROUTER_APP_TITLE), but the values are not derived/set consistently across runtimes and are easy to forget.

Goals

• Provide a single, predictable way to configure app identity metadata once and have it applied everywhere.
• Keep it opt-in/overrideable for local dev.
• Avoid hardcoding tenant-specific values in the repo.

Proposal

• Introduce a small metadata layer (e.g. src/common/metadata.py) that resolves:
  • app name/title
  • canonical origin URL / referer
  • optional app/environment identifiers
  • request id propagation
• Update LM configuration to apply these values to outbound OpenRouter/LiteLLM headers in a consistent way (Foundry + Railway).
• Document the env vars and recommended defaults in README + Foundry runbook.

Acceptance criteria

• In Foundry Compute Modules, with minimal config (ideally just APP_URL and APP_TITLE or similar), outbound LLM requests include the correct OpenRouter headers.
• In Railway, the same config keys work and produce the same headers.
• Clear logging at startup shows which metadata values were resolved (without leaking secrets).
• Existing behavior remains unchanged if no new metadata env vars are set.

Notes

This is primarily about observability/attribution and reducing configuration drift across deployment targets.