Docs API Reference SDK Reference Agentic Usage AI Skills

Contents

Responses API

An OpenAI Responses API-compatible endpoint for stateless multi-turn conversations, tool calling, and multimodal inputs. Ideal for agents and frameworks that use the OpenAI Python SDK ≥ 1.x with client.responses.create().

POST/v1/responses

Note

Accepts the same authentication and model routing as Chat Completions.

Request Body

modelrequired

string

Use an id or alias listed by GET /v1/models, such as "claude-sonnet-4.6" or "anthropic/claude-sonnet-4.6".

inputrequired

string | Item[]

User input — a plain string (single message) or an array of input items for multi-turn / multimodal conversations.

instructions

string

System-level instructions, equivalent to a system message. Must be re-sent on every request.

stream

boolean

If true, returns Responses API SSE stream events. Event types: response.created, response.output_text.delta, response.completed.

max_output_tokens

integer

Maximum number of output tokens to generate (includes reasoning tokens for o-series models).

temperature

number

Sampling values are forwarded without range validation. The model may clamp or ignore values outside its supported range.

top_p

number

Nucleus sampling probability mass. Default: 1

tools

Tool[]

Tool (function) definitions — same JSON Schema format as Chat Completions (accepts both the flat Responses shape and the nested shape). OpenAI's own built-in hosted tools (web_search_preview, file_search, computer_use_preview) are not supported; web search is available via plugins: [{id:"web"}] — supported only on some model routes.

tool_choice

string | object

Controls tool use: "auto", "none", or specific tool

parallel_tool_calls

boolean

Enable parallel function calling when tools are provided. Default: true. OpenAI-family models only — see note below.

response_format

object

Force structured JSON output. See Structured Output section

previous_response_id

string

This endpoint is stateless — passing a non-null value returns 400 (invalid_prompt) immediately, it is never accepted and ignored. Use stateless mode instead: pass the full conversation history in the input array.

Request Schema (TypeScript)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
type ResponsesRequest = {
  model: string;                    // Listed id or alias from GET /v1/models
  input: string | InputItem[];      // string or multi-turn array

  // Optional
  instructions?: string;            // System-level message
  stream?: boolean;                 // Default: false
  max_output_tokens?: number;
  temperature?: number;
  top_p?: number;
  tools?: Tool[];
  tool_choice?: "auto" | "none" | "required" | object;
  parallel_tool_calls?: boolean;    // Default: true
  previous_response_id?: string;    // Not supported — use full input array
};

type InputItem =
  | { type?: "message"; role: "user" | "assistant" | "system" | "developer"; content: string | ContentBlock[] }
  | { type: "function_call_output"; call_id: string; output: string }   // tool result
  | { type: "function_call"; call_id: string; name: string; arguments: string };

type ContentBlock =
  | { type: "input_text"; text: string }
  | { type: "input_image"; image_url: string; detail?: "auto" | "low" | "high" };

Example Request

1
2
3
4
5
6
7
8
curl https://api.bazaarlink.ai/v1/responses \
  -H "Authorization: Bearer $BAZAARLINK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-5.4-mini",
    "instructions": "You are a helpful assistant.",
    "input": "What is the capital of Taiwan?"
  }'

Response Format

BazaarLink returns a standardized response and removes upstream-specific fields. The opaque BazaarLink request id in the response matches x-request-id; use it with GET /v1/generation. Non-streaming responses also return the same id in x-generation-id. usage.cost is the USD amount charged to the customer. Stats may take a few seconds after completion; retry a 404 briefly.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
// Non-streaming response object
type ResponsesResponse = {
  id: string;             // Opaque BazaarLink request id; format may change
  object: "response";
  created_at: number;
  completed_at: number;
  status: "completed" | "failed" | "incomplete";
  model: string;
  output: OutputItem[];
  usage: {
    input_tokens: number;   // equivalent to prompt_tokens
    output_tokens: number;  // equivalent to completion_tokens
    total_tokens: number;
    cost?: number;          // USD amount charged to the customer for this request
  } | null;
  error: null | { code: string; message: string };
};

type OutputItem =
  | {
      type: "message";
      id: string;
      role: "assistant";
      status: "completed";
      content: Array<{ type: "output_text"; text: string; annotations: [] }>;
    }
  | { type: "function_call"; id: string; call_id: string; name: string; arguments: string; status: "completed" };

Migrating from Chat Completions

Replace messages with input (string or array), use instructions instead of a system-role message, and read output[0].content[0].text instead of choices[0].message.content.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
# Chat Completions (before)
response = client.chat.completions.create(
    model="openai/gpt-5.4-mini",
    messages=[
        {"role": "system", "content": "You are helpful."},
        {"role": "user",   "content": "Hello"},
    ]
)
text = response.choices[0].message.content

# Responses API (after)
response = client.responses.create(
    model="openai/gpt-5.4-mini",
    instructions="You are helpful.",
    input="Hello"
)
text = response.output[0].content[0].text

Limitations

previous_response_id or store: true is rejected with 400 (error code invalid_prompt) — not accepted and ignored. Always use stateless mode: pass the full conversation history in the input array.
OpenAI's own built-in hosted tools (web_search_preview, file_search, computer_use_preview) are not supported. Web search is available via plugins: [{id:"web"}] — supported only on some model routes.
background: true is accepted but ignored — every request runs synchronously to completion.