Module openai 

Modules

audio

Sources: local OpenAI create-operation docs; missing request/event schemas from openai/openai-python e12b81d3bbf644ec7045e152d69bc4b68d69cd48 and openai/openai-openapi 38170fdddbb6a1813eae6c6587ee17cf2987185b, checked 2026-09-13. Binary bodies and multipart streams remain transport-owned.

chat

codex

Codex-only account services, exposed under a Codex scoped mount. These are separate from the standard OpenAI API wire resources.

compact

Public Responses compaction body; source: Compact a response.md. Model is required-nullable, input optional-nullable; output is the same 28-item response union, with required usage accounting. Model resources remain open strings.

conversation

Conversation creation; source: Create Conversation.md. Response metadata is documented as arbitrary JSON, unlike the request metadata string map.

count_tokens

OpenAI POST /responses/input_tokens.

embeddings

OpenAI embeddings request/response wire shapes. Source: Create embeddings.md. Base64 wire vectors are also handled by the official SDK response parser: https://github.com/openai/openai-python/blob/main/src/openai/resources/embeddings.py

files

Sources checked 2026-09-13: openai/openai-python commit e12b81d3bbf644ec7045e152d69bc4b68d69cd48, src/openai/types/{file_object,file_create_params,file_list_params,file_deleted}.py; https://developers.openai.com/api/reference/resources/files/methods/list

guardian

Guardian and guardian-classifier use the Codex Responses client wire request. Sources: samples/codex/codex-rs/codex-api/src/common.rs (ResponsesApiRequest), endpoint/responses.rs (ResponsesEndpoint), and protocol/src/models.rs. Private client items differ from the public Responses schema. Headers such as x-openai-subagent belong to the outer WireRequest, not this body.

images

Sources: local OpenAI create-operation docs; missing request/event schemas from openai/openai-python e12b81d3bbf644ec7045e152d69bc4b68d69cd48 and openai/openai-openapi 38170fdddbb6a1813eae6c6587ee17cf2987185b, checked 2026-09-13. Binary bodies and multipart streams remain transport-owned.

memory

Codex memories/trace_summarize wire DTOs. Sources: codex-api/src/common.rs (MemorySummarizeInput/RawMemory/MemorySummarizeOutput) and endpoint/memories.rs. Trace items are arbitrary JSON by contract, not public Responses InputItems.

models

moderation

Standard OpenAI /v1/moderations, not Codex Guardian. Source: openai-python src/openai/types/{moderation_create_params,moderation_create_response, moderation,moderation_text_input_param,moderation_image_url_input_param}.py (verified 2026-10-01). Category keys remain open as upstream adds categories.

realtime

Realtime WebSocket event payloads and WebRTC setup forms. Client/server events are JSON values exchanged over an independent duplex connection. These DTOs neither open a connection nor frame/convert payloads.

rerank

OpenRouter /rerank and separately named DashScope /reranks wire shapes. Sources: openrouter/Submit a rerank request.md and dashscope/OpenAI-compatible Rerank.md. OpenRouter requires each result document; DashScope may omit it.

responses

Native OpenAI Responses wire shapes shared by Responses and input-token counting.

video

OpenRouter-shaped video generation extension in the OpenAI wire family. Sources: design/protocol-spec.md section 7 and local OpenRouter Submit/Poll/ Download video documents. Native* retains original OpenAI resource shapes; normalization/alias conversion is explicitly not performed by these DTOs. Native Video resource detail checked against openai-python commit e12b81d3bbf644ec7045e152d69bc4b68d69cd48, types/video.py (2026-09-13).

web_search

Native Codex /search DTOs; source codex-api/src/search.rs and endpoint/search.rs.