Responses API

The gateway also accepts the OpenAI Responses format at POST /v1/responses, so clients and agents built on it can call Ertas models.

resp = client.responses.create(
    model="ertas-ai/medertas-27b",
    input="List three red-flag symptoms of a headache.",
)
print(resp.output_text)
print(resp.usage)

What works today

• Basic API calls with input and instructions, with usage reported in the Responses format.
• Streaming with stream: true. The standard events arrive, ending with response.completed.
• Structured output with text.format and a JSON Schema.

Stateless calls only, for now

The gateway doesn't store responses yet, so each API call stands on its own. Stored responses are under development and coming soon. Until then:

• previous_response_id isn't supported. Send the earlier turns yourself in input, as you would with Chat Completions.
• Retrieving a response by ID (GET /v1/responses/{id}) isn't available.

history = [{"role": "user", "content": "A patient's BP is 156/94. Is that hypertension?"}]
first = client.responses.create(model="ertas-ai/medertas-27b", input=history)

history += [
    {"role": "assistant", "content": first.output_text},
    {"role": "user", "content": "What stage?"},
]
second = client.responses.create(model="ertas-ai/medertas-27b", input=history)

If an agent or framework relies on stored responses, use its Chat Completions mode instead, or turn off response storage in its settings.

For tool calls, use Chat Completions.