Responses

POST /v1/responses is the OpenAI Responses format. Codex CLI runs on it (wire_api = "responses", see Codex CLI).

Request#

curl https://ai-seller.vibe-codes.ru/v1/responses \
  -H "Authorization: Bearer $AISELLER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.6-sol",
    "instructions": "Answer briefly.",
    "input": "How many planets are in the Solar System?",
    "max_output_tokens": 200
  }'

from openai import OpenAI

client = OpenAI(base_url="https://ai-seller.vibe-codes.ru/v1", api_key="$AISELLER_API_KEY")
resp = client.responses.create(
    model="gpt-5.6-sol",
    instructions="Answer briefly.",
    input="How many planets are in the Solar System?",
    max_output_tokens=200,
)
print(resp.output_text)

Response#

{
  "id": "resp_...",
  "object": "response",
  "model": "gpt-5.6-sol",
  "status": "completed",
  "output": [
    {
      "type": "message",
      "role": "assistant",
      "content": [{"type": "output_text", "text": "Eight."}]
    }
  ],
  "usage": {"input_tokens": 20, "output_tokens": 2, "total_tokens": 22}
}

Specifics#

• No provider-side state. store is always false; previous_response_id, conversation, prompt and "background": true are rejected with 400 unsupported_parameter. Send the whole history in input of every request — Codex already does.
• Parameters have their own names: max_output_tokens instead of max_tokens, reasoning.effort instead of reasoning_effort, text.format instead of response_format. Chat Completions parameters are silently stripped here.
• Tools: function, custom, namespace, apply_patch, local_shell, computer, tool_search and shell with a local environment. The provider's built-in tools (web search, code execution, file search) are stripped.
• Streaming — "stream": true, response.* events; usage is in response.completed. See Streaming.

The full parameter list is in Request parameters.