Chat

POST /chat/completions takes the OpenAI chat completions body and returns an OpenAI chat completion. Use a model id from Models or "auto".

curl https://ewpire.com/api/v1/chat/completions \
  -H "Authorization: Bearer $EWPIRE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-5",
    "max_tokens": 400,
    "messages": [
      {"role": "system", "content": "You answer in two sentences."},
      {"role": "user", "content": "What is a vector database?"}
    ]
  }'

Streaming

Set "stream": true to get the answer as server-sent events: data: {chunk} lines, ending with data: [DONE]. Add "stream_options": {"include_usage": true} to get a last chunk with token usage and what the request cost.

curl -N https://ewpire.com/api/v1/chat/completions \
  -H "Authorization: Bearer $EWPIRE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "auto", "stream": true, "stream_options": {"include_usage": true}, "max_tokens": 300, "messages": [{"role": "user", "content": "Write a haiku about the sea."}]}'

Images in a message

Models with image in input_modalities read images. Pass them as image_url parts, as a link or a data: URL:

{
  "model": "auto",
  "messages": [{
    "role": "user",
    "content": [
      {"type": "text", "text": "What is in this picture?"},
      {"type": "image_url", "image_url": {"url": "https://example.com/photo.jpg"}}
    ]
  }]
}

With "auto", a message with an image goes to a model that can see it. A model that cannot read images answers 400 unsupported_content.

Parameters

top_p: 1, frequency_penalty: 0, presence_penalty: 0, logprobs: false, user and parallel_tool_calls are accepted and change nothing.

Where ewpire differs from OpenAI

• Unsupported parameters are refused, not ignored. stop, seed, logit_bias, an n above 1, any other value of the parameters above and any unknown parameter get 400 with the code unsupported_parameter and the name in param. You never believe a setting took effect when it did not.
• max_tokens defaults to 8,192. Without it an answer stops at 8,192 tokens.
• What a request cost comes in headers, or in the usage chunk when streaming. See Usage.
• Size limits. The request body can be up to 8 MB (images travel inline). The prompt is limited by the model's context, not by a character count.
• Retries are safe with Idempotency-Key. Send the header with a unique value per request; a repeat within 24 hours returns the first answer and is not charged again (it carries Idempotent-Replayed: true). The same key with a different body gets 422 idempotency_key_reused; a repeat while the first is still running gets 409 idempotency_key_in_use.