Generic checklist when moving from any OpenAI-compatible aggregator.

1. Confirm required endpoints

Ensure you only rely on:

  • POST /chat/completions
  • POST /embeddings
  • /files
  • POST /audio/speech
  • GET /models
  • (optional) Anthropic POST /messages, GET /anthropic/v1/models

2. Base URL and keys

  • OpenAI surface: https://51kik.com/v1
  • Anthropic surface: https://51kik.com/anthropic/v1
  • RouterBrain console keys

3. Model mapping

Fetch the new catalog; build old → new id map — slugs may differ.

curl -sS "https://51kik.com/v1/models" > models.json

4. Request bodies

  • Content parts (input_file, …)
  • Remove vendor-only fields — OpenAI differences
  • SSE proxy settings

5. Errors and retries

Parse per Errors; tune 403/429 handling.

6. Observability

Add x-trace-id (Request correlation).

7. Rollout

  1. Full regression before production cutover
  2. Canary or tenant-by-tenant cutover
  3. Watch error rate and latency

Related

Production checklist