Responses (OpenAI)

Responses requests, conversation state, and replies for the selected family.

Model protocol

POST /v1/responses is the main Codex interface for this connection. It supports text requests and history, instructions, client functions, non-streaming and streaming replies. Keys have no product modes; a server prompt is configured separately.

Supported contract

Input	Support
`input`	A string or message-item array with `input_text`; also `function_call` and `function_call_output` for continuing a client-tool loop
Generation	`max_output_tokens`, `stream`; application of `temperature` and `top_p` depends on the model and thinking
Client tools	`type: "function"` tools, `tool_choice`, and a compatible parallel tool loop; the client executes each function
Response	`output_text` and `function_call`; streaming ends with `response.completed`, while output-limit truncation returns `status: "incomplete"`

bash

curl https://api.guardrelay.ai/v1/responses \
  -H "Authorization: Bearer $GUARD_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-5-5",
    "instructions": "Answer briefly",
    "input": "Connection check: reply OK",
    "max_output_tokens": 64
  }'

Endpoint boundaries

PDF / input_file, input_image, native citations, and server-side Web/MCP tools are outside the Responses contract. Use Messages for PDFs and images.
top_k, explicit cache_control, native thinking/reasoning blocks, and structured text.format are outside the supported contract.
claude-fable-5 and claude-fable-5-1 are available here, including streaming and client functions. This endpoint retains its data-type limitations.
Explicitly forbidden types or semantics receive a local 400 before billing. Compatibility-only fields such as previous_response_id, include, truncation, or prompt_cache_key may be accepted without effect, so do not rely on them.

Kimi, Composer and Grok on this endpoint

All three families work here: Kimi, Composer and Grok. The surface is the same one; what is shorter is their list of accepted fields.

Kimi on this surface

Kimi is served here: the key and the address are the ordinary ones, model takes an exact Kimi id, and the text or the history goes into input. The ready Codex configuration for this surface is on the Codex page. If your client is happier with OpenAI Chat Completions, the same key works there too.

bash

curl -X POST https://api.guardrelay.ai/v1/responses \
  -H "Authorization: Bearer gd-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "k3",
    "input": "Hello!",
    "max_output_tokens": 128
  }'

Check Grok

bash

curl https://api.guardrelay.ai/v1/responses \
  -H "Authorization: Bearer gd-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "cursor-grok-4.6-low",
    "instructions": "Answer briefly",
    "input": "Connection check: reply OK",
    "max_output_tokens": 64
  }'

Check Composer

bash

curl https://api.guardrelay.ai/v1/responses \
  -H "Authorization: Bearer gd-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "composer-2.5",
    "instructions": "Answer briefly",
    "input": "Connection check: reply OK",
    "max_output_tokens": 64
  }'

What	How it works for Composer and Grok
Request fields	`model`, `input`, `instructions`, `max_output_tokens`, `stream`, `tools`
`input`	a string or an array of `type: "message"` items with roles `user`, `assistant`, `system`, `developer`
Content blocks	`input_text` and `output_text` with `type` and `text`; `cache_control` is accepted and ignored. Grok also accepts `input_image` with inline base64.
`function_call`, `function_call_output`	part of the client tool loop, served
`web_search_call`, `file_search_call`, `computer_call` and other provider-hosted items	rejected with `400 cursor_client_tools_unavailable`
`reasoning`, `temperature`, `top_p`, `store`, `include`, `truncation`, `metadata`, `previous_response_id`, `prompt_cache_key`, `thinking`	accepted and silently dropped
`parallel_tool_calls: true`, `service_tier` `auto`/`default`, `tool_choice: "auto"`	accepted at the default value only
`text`, any other value of the fields above	not accepted, `400 cursor_parameter_unavailable`
Response	`status: "completed"`, text in `output_text` and in `output[0].content[0].text`, cache in `usage.input_tokens_details.cached_tokens`
Response with a tool call	a `type: "function_call"` item is added to `output` with `call_id`, `name` and `arguments`; you answer it with a `function_call_output` item in the next request

Codex family Responses

All seven GPT models work through both aliases: POST https://api.guardrelay.ai/v1/responses and the retained POST https://api.guardrelay.ai/codex/v1/responses. They share one Responses lifecycle, one owner-scoped store, and the same billing.

bash

curl https://api.guardrelay.ai/codex/v1/responses \
  -H "Authorization: Bearer $GUARD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-6-astra",
    "instructions": "Answer briefly",
    "input": "Connection check: reply OK",
    "reasoning": {"effort": "medium"},
    "store": false
  }'

Responses supports JSON over HTTP, SSE with stream: true, and native WebSocket at wss://api.guardrelay.ai/v1/responses or wss://api.guardrelay.ai/codex/v1/responses. It supports instructions, reasoning effort, parallel client tools, images on compatible models, and actual usage; unknown fields are rejected before reserve or egress.

Confirmed Codex boundaries

store: true Responses, retrieve/delete, input items, and cancel use one owner-scoped lifecycle through both namespace aliases.
Conversations CRUD and their items use the same store and the same owner/key/payer rules through /v1 and /codex/v1.
Chat Completions and Messages pass the history supplied by the client. Guard does not invent server-side continuation across formats.
All seven models have an official maximum output of 128,000 tokens. Guard accepts and preserves the Responses max_output_tokens field as advisory. Forced truncation at 128K is not promised; billing follows actual tokens and the key's money rules.
For all seven Codex models, temperature is unavailable through the current connection. Omit the field entirely: both an explicit value and temperature: null return 400 invalid_request before reserve or egress. GPT-6 Sol and GPT-6 Luna still support reasoning.effort: none when temperature is absent. Supported efforts for the other models and the top_p contract are unchanged by this clarification.
Guard sends service_tier: "fast" upstream as priority; the response preserves the tier actually returned. The source may return default, so Fast availability and a speed-up are not guaranteed.
Batches, managed server agents, Projects/Tasks, server browser/MCP, and the broader hosted-resource API are outside this release.

GLM Responses

Use POST /v1/responses or POST /glm/v1/responses. input can be a string or a non-empty array of native input items.

bash

curl https://api.guardrelay.ai/glm/v1/responses \
  -H "Authorization: Bearer $GUARD_API_KEY" \
  -H "content-type: application/json" \
  -d '{"model":"glm-5.3","input":"Reply with OK","max_output_tokens":4096,"store":false}'

json

{
  "id": "resp_…",
  "model": "glm-5.3",
  "status": "completed",
  "output": [
    {"type": "reasoning", "content": []},
    {"type": "message", "content": [{"type": "output_text", "text": "OK"}]}
  ],
  "usage": {
    "input_tokens": 31,
    "input_tokens_details": {"cached_tokens": 0},
    "output_tokens": 12,
    "output_tokens_details": {"reasoning_tokens": 5}
  }
}

store and previous_response_id

Guard adds store: true when the field is omitted. To keep a response unavailable for continuation, send explicit store: false.

json

{
  "model": "glm-5.3",
  "previous_response_id": "resp_from_first_response",
  "input": "Continue from the previous answer",
  "max_output_tokens": 4096
}

Result	Meaning
`404 response_not_found`	The ID is unstored, unknown, or belongs to another key.
`409 response_account_unavailable`	The bound session is temporarily unavailable; state is not moved automatically.

Incomplete is a valid terminal

When the budget is exhausted, Responses returns status: "incomplete" and incomplete_details.reason: "max_output_tokens"; in SSE, response.incomplete is the terminal event. Completed work has exact usage and is billable.

Gemini Responses

Use POST /v1/responses with an exact Gemini model. input can be a string or an array of input items; without an explicit max_output_tokens the gateway uses 4096 tokens.

bash

curl https://api.guardrelay.ai/v1/responses \
  -H "Authorization: Bearer $GUARD_API_KEY" \
  -H "content-type: application/json" \
  -d '{"model":"gemini-4-argon","input":"Reply with OK","max_output_tokens":4096}'

Muse Responses

Use POST /v1/responses with an exact Muse model. input can be a string or an array of input items.

bash

curl https://api.guardrelay.ai/v1/responses \
  -H "Authorization: Bearer $GUARD_API_KEY" \
  -H "content-type: application/json" \
  -d '{"model":"muse-wonder-1.0","input":"Reply with OK","max_output_tokens":4096}'

Gemini continuation in Responses

previous_response_id continues a Gemini response that Guard saved for the key owner: any of the owner's keys continues the dialogue, and the response is kept for 30 days; without the store field the response is stored, and after a response with store: false there is nothing to continue. An unknown or expired id gets 404 gemini_history_not_found: then send the whole history in input. Gemini does not accept the conversation field: a request with it gets the error gemini_history_required.

Muse continuation in Responses

previous_response_id continues a Muse response created by the same key with store: true; without the store field the response is stored. After a response with store: false there is nothing to continue: send the history in input. An unknown id gets 404 response_not_found. Token counting needs the history in input, not in previous_response_id.

Responses (OpenAI)

Supported contract

Endpoint boundaries

Kimi, Composer and Grok on this endpoint

Kimi on this surface

Check Grok

Check Composer

Codex family Responses

Confirmed Codex boundaries

GLM Responses

store and previous_response_id

Incomplete is a valid terminal

Gemini Responses

Muse Responses

Gemini continuation in Responses

Muse continuation in Responses

Sections

On this page