Proxed.AI

Streaming Usage Finalization + Public Model Lists

We shipped a set of proxy + dashboard improvements focused on usage accuracy and SDK compatibility:

#
Streaming usage finalization

  • Streaming responses now update execution usage after the stream ends, so token counts, finish reasons, and costs are complete.
  • Provider-specific stream usage parsing for OpenAI, Anthropic, and Google is now consistent.
  • Added a toggle to disable storing stream summaries: set STREAM_STORE_SUMMARY=false.

#
Public model list endpoints

Model list endpoints are now public to match SDK discovery behavior (no auth header required):

  • GET /v1/openai/models
  • GET /v1/openai/{projectId}/models
  • GET /v1/anthropic/models
  • GET /v1/anthropic/{projectId}/models
  • GET /v1/google/models
  • GET /v1/google/{projectId}/models

These return Proxed-supported models with display names and pricing metadata.

#
Dashboard execution details

  • Execution tables and details now show model display names and pricing badges.
  • Filters and model selectors are more responsive with reduced recomputation.

Docs: https://docs.proxed.ai