MiroMind

Responses API

OpenAI Responses-style typed event streaming with reasoning + tool_call output items.

Overview

The Responses API delivers a richer event vocabulary than Chat Completions. Events are named (response.created, response.output_text.delta,response.completed, etc.) and arrive as SSE frames in event: <type>\ndata: <json>\n\n form, compatible with the OpenAI Python and TypeScript SDKs.

Base URL:https://api.miromind.ai
Auth:Bearer API Key
Pattern:Submit + StreamSubmit + Poll
Model Support

Each model declares which APIs it supports. Calling an unsupported endpoint returns 400 unsupported_api.

Model/v1/chat/completions/v1/responses
mirothinker-1-7-deepresearch✅✅
mirothinker-1-7-deepresearch-mini✅✅

Endpoints

POST/v1/responses
Submit Response

Create a response. With stream=true (default) the response body is an SSE event stream; with stream=false the connection blocks until the workflow finishes and returns the full Response JSON. With background=true the call returns immediately and you fetch the result via GET /v1/responses/:id.

Headers

AuthorizationRequired

Your MiroMind API key

Bearer YOUR_API_KEY
Content-TypeRequired

Must be application/json

application/json

Parameters

modelstringRequired

Model ID, e.g. "mirothinker-1-7-deepresearch-mini".

inputstring | arrayRequired

A user prompt as a plain string, or an array of {type, role, content} input items.

streamboolean

Whether to stream events via SSE. Defaults to true.

backgroundboolean

If true, returns the Response object immediately (status="in_progress") without holding a connection. Use GET /v1/responses/:id (or ?stream=true to resume the SSE stream) to fetch the result later. Mutually exclusive with stream=true.

max_output_tokensinteger

Maximum number of tokens to generate.

temperaturenumber

Sampling temperature.

metadataobject

Free-form key-value metadata stored with the response.

Request

curl -N -X POST https://api.miromind.ai/v1/responses \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mirothinker-1-7-deepresearch-mini",
    "input": "Summarize the recent advances in fusion energy."
  }'

Response

event: response.created
data: {"type":"response.created","sequence_number":1,"response":{"id":"resp_<id>","status":"in_progress",...}}

event: response.in_progress
data: {"type":"response.in_progress","sequence_number":2,...}

event: response.output_item.added
data: {"type":"response.output_item.added","sequence_number":3,"output_index":0,"item":{"id":"rs_<id>","type":"reasoning","summary":[]}}

event: response.reasoning_text.delta
data: {"type":"response.reasoning_text.delta","sequence_number":4,"item_id":"rs_<id>","output_index":0,"content_index":0,"delta":"Thinking about ..."}

... (more reasoning + optional tool_call output items) ...

event: response.output_item.added
data: {"type":"response.output_item.added","sequence_number":N,"output_index":1,"item":{"id":"msg_<id>","type":"message","role":"assistant","content":[]}}

event: response.output_text.delta
data: {"type":"response.output_text.delta","sequence_number":N+1,"item_id":"msg_<id>","output_index":1,"content_index":0,"delta":"Recent advances ..."}

event: response.output_text.done
data: {"type":"response.output_text.done","sequence_number":N+M,...}

event: response.completed
data: {"type":"response.completed","sequence_number":N+M+1,"response":{"id":"resp_<id>","status":"completed","output":[...],"usage":{...}}}
GET/v1/responses/:id
Get / Resume Stream

Fetch a Response by id. Default: returns the full Response object snapshot reconstructed from the workflow's event stream. With ?stream=true: opens an SSE stream that replays the entire event history through the translator and continues live until the workflow terminates — clients that lost their connection mid-stream can reconnect with ?stream=true&after=<seq> to skip events they already saw.

Headers

AuthorizationRequired

Your MiroMind API key

Bearer YOUR_API_KEY

Parameters

streamboolean

If true, returns SSE event stream (replay + live). If absent or false, returns the JSON snapshot.

afterinteger

Only meaningful with stream=true. Skip events whose sequence_number is ≤ this value. Use to resume a stream where you left off.

Request

curl https://api.miromind.ai/v1/responses/resp_<id> \
  -H "Authorization: Bearer YOUR_API_KEY"

Response

Snapshot
{
  "id": "resp_<workflow_id>",
  "object": "response",
  "status": "completed",
  "model": "mirothinker-1-7-deepresearch-mini",
  "output": [...],
  "usage": {...}
}
GET/v1/responses
List Responses

List the authenticated user's recent responses. Lifecycle metadata only — no output[]; fetch individual responses via GET /v1/responses/:id.

Headers

AuthorizationRequired

Your MiroMind API key

Bearer YOUR_API_KEY

Parameters

limitinteger

Default 20, max 100.

Request

cURL
curl https://api.miromind.ai/v1/responses?limit=20 \
  -H "Authorization: Bearer YOUR_API_KEY"

Response

Response
{
  "object": "list",
  "data": [
    {
      "id": "resp_<workflow_id>",
      "object": "response",
      "status": "completed",
      "model": "mirothinker-1-7-deepresearch-mini",
      "created_at": 1700000000,
      "started_at": 1700000001,
      "completed_at": 1700000020
    }
  ],
  "has_more": true
}
POST/v1/responses/:id/cancel
Cancel Response

Cancel an in-progress response. Idempotent — cancelling an already-terminal response is a no-op.

Headers

AuthorizationRequired

Your MiroMind API key

Bearer YOUR_API_KEY

Request

cURL
curl -X POST https://api.miromind.ai/v1/responses/resp_<id>/cancel \
  -H "Authorization: Bearer YOUR_API_KEY"

Response

Response
{
  "id": "resp_<workflow_id>",
  "object": "response",
  "status": "cancelled"
}
Streaming Format

SSE frames use OpenAI Responses convention:

event: <event_type>
data: <json_payload>

Each payload includes type (matches the event line) and sequence_number (monotonic int starting at 1). Item-scoped events also include item_id / output_index / content_index.

The stream ends with response.completed (or response.failed). There is no [DONE] sentinel — the named events are self-terminating.

Output Item Shape

The final response.output[] array contains the ordered output items: planning text alternates with tool calls, then a final message:

output[0]  reasoning   — first planning turn
output[1]  tool_call   — { name, arguments, result, status: "completed" }
output[2]  reasoning   — continued planning
output[3]  tool_call   — ...
output[k]  message     — final answer text

Reasoning items (type: "reasoning") carry the model's planning narrative. Streamed via response.reasoning_text.delta.

Tool-call items (type: "tool_call") describe an internal tool the agent invoked while researching. Each carries { id, type: "tool_call", name, arguments, result, status }. The tool already ran on the server — the call is informational. Two events fire: response.output_item.added at invocation (with status: "in_progress") and response.output_item.done at completion (with status: "completed" and the result).

Important — tool_call vs OpenAI standard function_call: We deliberately use type: "tool_call", not OpenAI's type: "function_call". Our tools run on the server, so clients should NOT execute them locally or submit a function_call_output back. The OpenAI Python and TypeScript SDKs treat tool_call as a generic / unknown output item and pass it through unchanged — they do not trigger their function-execution flow. Read item.result directly.

Message item (type: "message") is the final user-visible answer. content[0].text holds the text. Streamed via response.output_text.delta.

MiroMindAI Platform - Build Together. Evolve Forever.