Create a response

Generate a response from a user message. This is the single conversational endpoint on EU GPT.

Endpoint#

POST /v1/responses
Authorization: Bearer eugpt_<token>
Content-Type: application/json

The full interactive reference is at API reference. This page documents the request and response shapes in prose so you can read top-to-bottom.

Request body#

{
  model?: string,                  // default: "auto"
  input: string | ContentItem[],   // required
  stream?: boolean,                // default: true
  instructions?: string | null,
  conversation_id?: string | null, // UUID
  project_id?: string | null,      // UUID
}

model#

Accepted for OpenAI-SDK compatibility but ignored — the router always selects the model. Pass "auto" (the default and only meaningful value); any concrete id you send has no effect. The model actually used is reported in the response model field. See Models for routing behaviour.

"model": "auto"

input#

The user message. Either a plain string for text-only input, or a list of ContentItems for structured input.

String form:

"input": "Summarise the Q3 audit in three bullets."

Structured form:

"input": [
  {
    "role": "user",
    "content": [
      { "type": "input_text", "text": "Summarise the Q3 audit in three bullets." }
    ]
  }
]

Each ContentItem has:

• role (optional) — usually "user". If omitted, treated as user.
• content — a list of InputContent parts, each with type: "input_text" and a text field.

stream#

Boolean. Defaults to true. When true, the response is Server-Sent Events. When false, the response is a single JSON document.

"stream": false

See Streaming for the trade-offs and Streaming events for the event taxonomy.

instructions#

System prompt for this response. Steers tone, format, persona, language.

"instructions": "Always respond in Dutch. Cite sources when using web_search."

If conversation_id refers to a brand-new conversation, the instructions are also stored as the conversation’s system message. For existing conversations, they apply to this response only.

conversation_id#

UUID of an existing conversation. Append to it. The conversation must be owned by the user that issued the API key.

"conversation_id": "8f14e45f-ceea-467a-a4ed-a9e9a5cb16ee"

Omit (or pass null) for a stateless request. See Conversations.

project_id#

UUID of a project for RAG retrieval. The project’s file corpus is searched and relevant chunks are added to the prompt context.

"project_id": "1c8b9a7f-2d3e-4f5a-9b8c-7d6e5f4a3b2c"

The caller must be a member of the project. Otherwise the response is 403.

Response — non-streaming (stream: false)#

{
  id: string,             // UUID
  object: "response",
  created_at: number,     // Unix epoch ms
  status: "completed",
  model: string,          // resolved concrete model id
  output: ContentItem[],  // structured output (tool calls, content parts, …)
  output_text: string,    // flat concatenated text
}

• output_text is the convenience field. For plain-text consumers this is all you need.
• output carries the structured items: text content, tool calls and their outputs, file references. Match the shape of the streamed response.output_item.added events.

status is always completed when a response body is returned — failures come back as HTTP error responses with the error envelope, not a status field.

Response — streaming (stream: true)#

Content-Type: text/event-stream. Each event has the shape:

event: message
data: { "type": "<event-name>", "sequence_number": <int>, … }

The full event reference is at Streaming events. The minimum sequence is:

response.created
response.output_text.delta  (one or more)
response.completed

The terminal event is always response.completed (success) or error (failure mid-stream).

Examples#

Minimal stateless string input#

curl https://chat.eugpt.ai/v1/responses \
  -H "Authorization: Bearer $EUGPT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input":"Hello, world.","stream":false}'

Multi-turn conversation#

curl https://chat.eugpt.ai/v1/responses \
  -H "Authorization: Bearer $EUGPT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "auto",
    "input": "And what was the previous answer in one sentence?",
    "conversation_id": "8f14e45f-ceea-467a-a4ed-a9e9a5cb16ee",
    "stream": false
  }'

Project-scoped RAG#

curl https://chat.eugpt.ai/v1/responses \
  -H "Authorization: Bearer $EUGPT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "input": "What does the 2024 audit say about server costs?",
    "project_id": "1c8b9a7f-2d3e-4f5a-9b8c-7d6e5f4a3b2c",
    "stream": false
  }'