.. _envoy_v3_api_file_envoy/type/ai/v3/api_protocol.proto:

AI API protocol (proto)
=======================






.. _envoy_v3_api_enum_type.ai.v3.ApiProtocol:

Enum type.ai.v3.ApiProtocol
---------------------------


:repo:`[type.ai.v3.ApiProtocol proto] <api/envoy/type/ai/v3/api_protocol.proto#L24>`

Identifies the AI API contract spoken on the wire.

This names request/response schemas and streaming event semantics, not the
organization operating the backend: OpenAI, Azure OpenAI, vLLM, and any
other OpenAI-compatible backend all speak ``OPENAI_CHAT_COMPLETIONS``.
Provider identity, when needed, must come from configuration or routing
metadata.

Response framing is separate: JSON, SSE, and streamed JSON-array variants
of the same API share one ``ApiProtocol`` value.



.. _envoy_v3_api_enum_value_type.ai.v3.ApiProtocol.API_PROTOCOL_UNSPECIFIED:


API_PROTOCOL_UNSPECIFIED
  *(DEFAULT)* ⁣

.. _envoy_v3_api_enum_value_type.ai.v3.ApiProtocol.OPENAI_CHAT_COMPLETIONS:


OPENAI_CHAT_COMPLETIONS
  ⁣OpenAI-compatible ``/v1/chat/completions`` API.


.. _envoy_v3_api_enum_value_type.ai.v3.ApiProtocol.OPENAI_RESPONSES:


OPENAI_RESPONSES
  ⁣OpenAI-compatible ``/v1/responses`` API.


.. _envoy_v3_api_enum_value_type.ai.v3.ApiProtocol.ANTHROPIC_MESSAGES:


ANTHROPIC_MESSAGES
  ⁣Anthropic-compatible ``/v1/messages`` API.


.. _envoy_v3_api_enum_value_type.ai.v3.ApiProtocol.GEMINI_GENERATE_CONTENT:


GEMINI_GENERATE_CONTENT
  ⁣Gemini ``generateContent`` / ``streamGenerateContent`` APIs.