Responses

The Responses API is a stateful way to run multi-turn conversations. Instead of resending the whole conversation on every turn, you send only the new input and reference the previous turn by its id — EcoLink keeps the conversation on the server. For reasoning models, the model's thinking is carried across turns as well, which improves quality on follow-up questions.

It is available for text-generation platform models. For a single, self-contained request, or for embeddings, vision, audio, or your own inference instances, use Chat completions instead.

POST https://api.ecohash.com/v1/responses
GET  https://api.ecohash.com/v1/responses/{id}
POST https://api.ecohash.com/v1/responses/{id}/cancel

Headers

Header	Value
Authorization	Bearer eco_YOUR_KEY
Content-Type	application/json

Request body

Field	Type	Required	Notes
model	string	yes	A text-generation platform model, e.g. qwen3.6-27b
input	string or array	yes	The new user input for this turn — a plain string, or an array of typed content items
instructions	string	no	System-level guidance applied to this response
previous_response_id	string	no	The id of the response this turn continues. Omit to start a new conversation
max_output_tokens	integer	no	Upper bound on the generated length
temperature	number	no	0–2, default 1.0
top_p	number	no	0–1
stream	bool	no	true for SSE streaming

Create a response

curl https://api.ecohash.com/v1/responses \
  -H "Authorization: Bearer eco_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.6-27b",
    "instructions": "You are a concise assistant.",
    "input": "What is EcoLink?"
  }'

Response:

{
  "id": "resp_abc123",
  "object": "response",
  "created_at": 1776391234,
  "model": "qwen3.6-27b",
  "status": "completed",
  "output": [{
    "type": "message",
    "role": "assistant",
    "content": [{"type": "output_text", "text": "EcoLink is a GPU cloud platform ..."}]
  }],
  "usage": {
    "input_tokens": 18,
    "output_tokens": 24,
    "total_tokens": 42
  }
}

The usage block uses input_tokens and output_tokens (billed the same as prompt_tokens / completion_tokens on chat completions). When a prior turn's context is reused, input_tokens_details.cached_tokens reports the cached portion, which is billed at the lower cache-hit rate.

Continue the conversation

Send the next turn with previous_response_id set to the prior id. You send only the new input — the earlier turns are retained on the server.

curl https://api.ecohash.com/v1/responses \
  -H "Authorization: Bearer eco_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.6-27b",
    "previous_response_id": "resp_abc123",
    "input": "Which GPUs does it offer?"
  }'

Keep model the same across a conversation. Continue from the most recent response id you received.

Retrieve a response

curl https://api.ecohash.com/v1/responses/resp_abc123 \
  -H "Authorization: Bearer eco_YOUR_KEY"

Returns the response object in the same shape as the create call.

Cancel a response

curl -X POST https://api.ecohash.com/v1/responses/resp_abc123/cancel \
  -H "Authorization: Bearer eco_YOUR_KEY"

Streaming

With "stream": true, the response is a sequence of data: <json>\n\n server-sent events. Each event has a type; text arrives on response.output_text.delta events, and the final response.completed event carries the full response object including usage:

data: {"type":"response.created","response":{"id":"resp_abc123","status":"in_progress"}}

data: {"type":"response.output_text.delta","delta":"Eco"}

data: {"type":"response.output_text.delta","delta":"Link"}

data: {"type":"response.completed","response":{"id":"resp_abc123","status":"completed","usage":{"input_tokens":18,"output_tokens":24,"total_tokens":42}}}

data: [DONE]

Response headers

Header	Meaning
x-ecolink-region	Region that served this request
x-ecolink-request-id	Unique ID for this request (include in support requests)

Error responses

See Errors for the full list. Common:

HTTP	Meaning
400	Malformed body, or a model that does not support this endpoint
401	Bad or missing API key
402	Balance insufficient or model not priced
404	Unknown model ID, or a previous_response_id / response id that is no longer available
429	Rate limited
503	No healthy region for this model

A retained conversation is available for 24 hours after its most recent turn. After that, continue from a fresh response with no previous_response_id.