Generate a response using the OpenAI Responses API format. This endpoint is an alternative to /v1/chat/completions with a different request/response shape.

⚠️ Known limitation

Tool calling is not yet supported on this endpoint. If you need tool calling, use /v1/chat/completions or /v1/messages instead.

Endpoint

POST https://api.runars.ca/v1/responses

Authentication

Include your API key as a Bearer token:

-H "Authorization: Bearer sk-runars-..."

Request body

For a comparison of common parameters across all endpoints, see Parameters Reference.

The Responses API uses input and instructions instead of the messages format:

Field	Type	Required	Description
`model`	string	Yes	Model identifier, must be `"glm-5.2"`
`input`	string	Yes	The user's input/prompt
`instructions`	string	No	System instructions (similar to system prompt)
`max_tokens`	integer	No	Maximum tokens in the response
`stream`	boolean	No	Enable streaming (default: false)

Response (non-streaming)

{
  "id": "response_...",
  "object": "response",
  "created": 1234567890,
  "model": "glm-5.2",
  "output_text": "The response text here.",
  "status": "completed",
  "finish_reason": "stop"
}

Response (streaming)

With stream: true, the endpoint returns server-sent events:

event: response.created
data: {"id":"response_...","object":"response","created":1234567890,"model":"glm-5.2"}

event: response.output_text.delta
data: {"object":"response.delta","delta":{"output_text":"The "}}

event: response.output_text.delta
data: {"object":"response.delta","delta":{"output_text":"response "}}

event: response.completed
data: {"object":"response.completed","id":"response_...","status":"completed","finish_reason":"stop"}

Examples

Simple text request

curl https://api.runars.ca/v1/responses \
  -H "Authorization: Bearer sk-runars-..." \
  -H "content-type: application/json" \
  -d '{
    "model": "glm-5.2",
    "input": "What is the capital of France?"
  }'

With instructions

curl https://api.runars.ca/v1/responses \
  -H "Authorization: Bearer sk-runars-..." \
  -H "content-type: application/json" \
  -d '{
    "model": "glm-5.2",
    "instructions": "You are a helpful assistant. Keep responses concise.",
    "input": "Explain quantum computing in one sentence."
  }'

Using the OpenAI SDK

import OpenAI from "openai"

const openai = new OpenAI({
  baseURL: "https://api.runars.ca/v1",
  apiKey: process.env.OPENAI_API_KEY,
})

const response = await openai.beta.responses.create({
  model: "glm-5.2",
  input: "Tell me a short story about a robot.",
  max_tokens: 500,
})

console.log(response.output_text)

Streaming

const stream = await openai.beta.responses.create({
  model: "glm-5.2",
  input: "Tell me a short story about a robot.",
  stream: true,
})

for await (const event of stream) {
  if (event.object === "response.output_text.delta") {
    process.stdout.write(event.delta.output_text)
  }
}

Web search

Add OpenAI's web search tool and Runars executes the searches gateway-side:

{
  "model": "glm-5.2",
  "input": "What happened in the news today?",
  "tools": [
    {
      "type": "web_search",
      "search_context_size": "medium",
      "filters": { "allowed_domains": ["example.com"] }
    }
  ]
}

Output: web_search_call items (action.type: "search" with the query) precede the message item; the message's output_text carries url_citation annotations for the sources used.
Usage: usage.web_search_requests reports the executed-search count.
Options: search_context_size (low/medium/high → how many results the model sees), filters.allowed_domains / blocked_domains, and user_location are honored. web_search_preview is accepted as an alias.
Billing: $10 / 1,000 executed searches (standard rate) plus normal token costs; errored searches are not billed. Rounds of web_search_call items in follow-up input are accepted and skipped.

Comparison to `/v1/chat/completions`

Aspect	Chat Completions	Responses
Request format	`messages` array	`input` string + `instructions`
Response format	`choices[0].message.content`	`output_text`
Multi-turn	✅ (via message history)	⚠️ (requires manual state)
Tool calling	✅	❌
Streaming events	`chat.completion.chunk`	`response.*` events

Use /v1/responses if:

You prefer the simpler input/output_text API shape
You're porting code from a different API that uses this format
You don't need tool calling

Use /v1/chat/completions if:

You need tool calling
You're building multi-turn conversations
You're using the OpenAI SDK

Error responses

400 Bad Request

Missing required field:

{
  "error": {
    "message": "Invalid request: missing required field 'input'",
    "type": "invalid_request_error"
  }
}

401 Unauthorized

Invalid or missing API key:

{
  "error": {
    "message": "Unauthorized",
    "type": "invalid_request_error",
    "code": "invalid_api_key"
  }
}

429 Too Many Requests

Rate limit exceeded:

{
  "error": {
    "message": "Rate limit exceeded",
    "type": "server_error",
    "code": "insufficient_quota"
  }
}

Future roadmap

Tool calling support is planned for this endpoint. In the meantime, use /v1/chat/completions or /v1/messages if you need tools.

POST /v1/responses