Responses
POST /v1/responses· OpenAI Responses 兼容(按请求形状分流)
/v1/responses 主要服务于 Codex 等只说 Responses API 的客户端。网关按请求形状
判断来源:
- Codex 形状(带
instructions/tools/reasoning/input消息数组)→ 内部 翻译成 Chat Completions 执行,再把结果翻译回object: "response"形状。 - 普通文本形状(
input为字符串)→ 返回 OpenAIchat.completion形状(与/v1/chat/completions一致),方便非 Codex 调用方复用。
请求#
POST /v1/responses
Authorization: Bearer nx-...
Content-Type: application/jsonBody(通用)#
| 字段 | 类型 | 必填 | 说明 |
|---|---|---|---|
model |
string | — | 默认 nexith(映射到 nexith-core) |
input |
string | array | ✅ | 字符串,或 Responses input item 数组 |
stream |
bool | — | 默认 false |
max_output_tokens |
integer | — | 推理 + 正文的总预算;过小会导致正文为空 |
previous_messages |
array | — | 可选的多轮历史(Nexith 扩展,Chat 形状) |
Codex 形状额外字段#
| 字段 | 说明 |
|---|---|
instructions |
顶层指令,翻译为 system 消息 |
tools |
Responses 工具定义({type:"function", name, parameters}),翻译为 Chat tools |
tool_choice |
auto / none / required / 指定函数 |
reasoning |
{effort: ...},翻译为 reasoning_effort |
input items#
[
{"type": "message", "role": "user", "content": [{"type": "input_text", "text": "hello"}]},
{"type": "message", "role": "user", "content": [{"type": "input_image", "image_url": "https://…"}]},
{"type": "function_call_output", "call_id": "call_1", "output": "42"}
]支持 input_text、input_image、message、function_call、function_call_output。
响应(Codex 形状)#
{
"id": "resp_nx0000000000000001",
"object": "response",
"created_at": 1795055384,
"model": "nexith-core",
"status": "completed",
"output": [
{
"type": "message",
"id": "msg_nx0000000000000000",
"role": "assistant",
"status": "completed",
"content": [{"type": "output_text", "text": "ok", "annotations": []}]
}
],
"usage": {"input_tokens": 324, "output_tokens": 32, "total_tokens": 356}
}status:completed/incomplete(如达到max_output_tokens)。- 工具调用输出为
function_callitem。
响应(普通文本形状)#
与 Chat Completions 完全一致(object: "chat.completion")。
流式#
stream=true 返回 SSE。Codex 形状的流会输出 Responses 事件;文本形状的流为
chat.completion.chunk。帧细节见 流式协议。
示例#
curl https://api.nexith.ai/v1/responses \
-H "Authorization: Bearer $NEXITH_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "nexith-core",
"instructions": "You are a helpful coding agent.",
"input": [{"type":"message","role":"user","content":[{"type":"input_text","text":"say ok"}]}],
"max_output_tokens": 256
}'
max_output_tokens包含隐藏推理。推理模型下建议 ≥ 512;设得太小会出现status: "incomplete"且正文为空。
客户端配置#
Codex CLI#
# ~/.codex/config.toml
[model_providers.nexith]
name = "Nexith"
base_url = "https://api.nexith.ai/v1"
env_key = "NEXITH_API_KEY"
wire_api = "responses"详见 Codex 接入。