Skip to main content
Use this guide when Helicone sits between your application and an AI provider. The main request change is usually small, but you should plan the observability cutover as carefully as the endpoint change.

What changes

1

List the Helicone behavior you rely on

Record custom properties, sessions, user identifiers, caching, rate limits, alerts, and dashboards that affect production operations. Do not assume proxy removal automatically recreates those workflows elsewhere.
2

Replace the proxy configuration

Point your OpenAI-compatible client at Phaseo and remove Helicone-Auth and other Helicone-specific headers.
3

Move request visibility

Use Phaseo activity, request IDs, provider attempts, and pricing details for operational debugging. If your team exports telemetry elsewhere, connect the matching Observability integration before the cutover.
4

Test and roll out

Verify that requests complete without Helicone headers and that your team can still answer the operational questions it uses during incidents. Move traffic gradually once both inference and observability checks pass.

Migration checklist

  • The Helicone proxy URL and authentication headers have been removed.
  • Production model IDs are verified against GET /v1/models.
  • Required request metadata remains available in Phaseo or your telemetry stack.
  • Alerts and dashboards have an intentional replacement where needed.
  • Rollback remains a configuration-only endpoint and key change.

Next steps

Last modified on July 26, 2026