[GH-ISSUE #245] Feature Request: Add OpenAI Responses API Compatibility for Codex Desktop #90

Open
opened 2026-10-26 09:45:13 +00:00 by bsakaguchi · 2 comments
Owner

Originally created by @Start-Ten on GitHub (Jun 23, 2026).
Original GitHub issue: https://github.com/deepseek-ai/awesome-deepseek-agent/issues/245

Feature Request: Add OpenAI Responses API (/v1/responses) Compatibility

Background

OpenAI has introduced the new Responses API (/v1/responses) as the evolution of the Chat Completions API, with built-in support for powerful tools (code interpreter, image generation, web search) and structured outputs. OpenAI's Codex Desktop natively uses the Responses API as its communication protocol.

DeepSeek API already supports the OpenAI-compatible /v1/chat/completions endpoint, but does not yet support the /v1/responses endpoint. This prevents tools like Codex Desktop from connecting directly to DeepSeek models.

Community solutions like codex-proxy exist to bridge this gap via protocol translation, but they introduce additional latency, complexity, and maintenance overhead.

Suggestion

Add OpenAI Responses API compatible endpoints to DeepSeek's API service:

POST https://api.deepseek.com/v1/responses
POST https://api.deepseek.com/v1/responses/compact

Support the standard Responses API request/response format, including:

  • input field (replacing messages)
  • Built-in tools (e.g., code_interpreter, web_search)
  • response_format for structured outputs
  • SSE streaming responses

Why This Matters

  1. Native Codex Desktop Support — Codex uses Responses API natively. With compatibility, users can point Codex directly to DeepSeek's API endpoint in settings
  2. Eliminate Middleware — No need for third-party proxy services like codex-proxy, reducing latency and complexity
  3. Keep Pace with Industry Standards — Responses API is OpenAI's new standard protocol, and more tools will adopt it going forward
  4. Unlock DeepSeek V4 Flash's Potential — Fast inference + native tool calling is a powerful combination for coding workflows

Reference

Originally created by @Start-Ten on GitHub (Jun 23, 2026). Original GitHub issue: https://github.com/deepseek-ai/awesome-deepseek-agent/issues/245 ## Feature Request: Add OpenAI Responses API (/v1/responses) Compatibility ### Background OpenAI has introduced the new **Responses API** (`/v1/responses`) as the evolution of the Chat Completions API, with built-in support for powerful tools (code interpreter, image generation, web search) and structured outputs. OpenAI's **Codex Desktop** natively uses the Responses API as its communication protocol. DeepSeek API already supports the OpenAI-compatible `/v1/chat/completions` endpoint, but does **not** yet support the `/v1/responses` endpoint. This prevents tools like Codex Desktop from connecting directly to DeepSeek models. Community solutions like [codex-proxy](https://github.com/icebear0828/codex-proxy) exist to bridge this gap via protocol translation, but they introduce additional latency, complexity, and maintenance overhead. ### Suggestion Add OpenAI Responses API compatible endpoints to DeepSeek's API service: ``` POST https://api.deepseek.com/v1/responses POST https://api.deepseek.com/v1/responses/compact ``` Support the standard Responses API request/response format, including: - `input` field (replacing `messages`) - Built-in `tools` (e.g., `code_interpreter`, `web_search`) - `response_format` for structured outputs - SSE streaming responses ### Why This Matters 1. **Native Codex Desktop Support** — Codex uses Responses API natively. With compatibility, users can point Codex directly to DeepSeek's API endpoint in settings 2. **Eliminate Middleware** — No need for third-party proxy services like codex-proxy, reducing latency and complexity 3. **Keep Pace with Industry Standards** — Responses API is OpenAI's new standard protocol, and more tools will adopt it going forward 4. **Unlock DeepSeek V4 Flash's Potential** — Fast inference + native tool calling is a powerful combination for coding workflows ### Reference - Codex Proxy project (https://github.com/icebear0828/codex-proxy) implements Chat Completions to Responses API protocol translation, serving as a reference implementation - OpenAI Responses API docs: https://platform.openai.com/docs/guides/responses-api
Author
Owner

@tianhao8687 commented on GitHub (Aug 17, 2026):

Update (August 2026): the core endpoint requested here has shipped. DeepSeek now documents native Responses API support at https://api.deepseek.com, and the current Codex integration guide configures Codex directly with wire_api = "responses"—no Chat Completions translation proxy is required for the supported path.

The compatibility is intentionally partial, so this issue may be worth closing for endpoint availability and tracking parity gaps separately. The current Responses API compatibility table says:

  • deepseek-v4-flash and deepseek-v4-pro are supported;
  • SSE streaming, function tools, server-side web search, structured text format, and Codex's apply_patch custom tool are supported;
  • file_search, code_interpreter, computer_use, MCP, and most other built-in tool types are ignored;
  • previous_response_id, conversations, storage, and background mode are not supported because the endpoint is stateless;
  • unknown input-item types are ignored, which is relevant to newer Codex collaboration items such as agent_message.

I do not see /v1/responses/compact documented as a supported endpoint. So “native Codex connectivity” is now available, while “full OpenAI Responses feature parity” is not yet complete.

<!-- gh-comment-id:5317549665 --> @tianhao8687 commented on GitHub (Aug 17, 2026): Update (August 2026): the core endpoint requested here has shipped. DeepSeek now documents native Responses API support at `https://api.deepseek.com`, and the current [Codex integration guide](https://api-docs.deepseek.com/quick_start/agent_integrations/codex) configures Codex directly with `wire_api = "responses"`—no Chat Completions translation proxy is required for the supported path. The compatibility is intentionally partial, so this issue may be worth closing for endpoint availability and tracking parity gaps separately. The current [Responses API compatibility table](https://api-docs.deepseek.com/guides/responses_api) says: - `deepseek-v4-flash` and `deepseek-v4-pro` are supported; - SSE streaming, function tools, server-side web search, structured text format, and Codex's `apply_patch` custom tool are supported; - `file_search`, `code_interpreter`, `computer_use`, MCP, and most other built-in tool types are ignored; - `previous_response_id`, conversations, storage, and background mode are not supported because the endpoint is stateless; - unknown input-item types are ignored, which is relevant to newer Codex collaboration items such as `agent_message`. I do not see `/v1/responses/compact` documented as a supported endpoint. So “native Codex connectivity” is now available, while “full OpenAI Responses feature parity” is not yet complete.
Author
Owner

@gomezedelce9-design commented on GitHub (Aug 20, 2026):

<!-- gh-comment-id:5361678347 --> @gomezedelce9-design commented on GitHub (Aug 20, 2026): <!-- Failed to upload "Facebook_1786893211527(360p).mp4" -->
Sign in to join this conversation.
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
deepseek-ai/awesome-deepseek-agent#90
No description provided.