Chat Completions

POST https://api.vikey.ai/v1/chat/completions

The request and response follow OpenAI Chat Completions. This endpoint serves every text model on Vikey, not only OpenAI models.

Example

curl https://api.vikey.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-5.5",
    "max_tokens": 1024,
    "messages": [
      {"role": "system", "content": "Answer briefly."},
      {"role": "user", "content": "What is the difference between let and const in JavaScript?"}
    ]
  }'

Common parameters

Parameter	Notes
model	Required. A model id, such as openai/gpt-6.1-sol or vikey/agent-gpt.
messages	Required. Messages with a role (system, user, assistant, tool) and content.
max_tokens	Output token limit. Also used for the balance check, so keep it sensible.
stream	true for Server-Sent Events. The stream ends with data: [DONE].
tools, tool_choice	Function calling in the OpenAI format.
temperature, top_p, stop	Passed to the model.
response_format	JSON mode or JSON schema, for models that support it.

Other parameters are passed to the model as they are. A model that does not support a parameter may ignore it or return an error.

Images (vision)

For models with vision, send images as part of content:

{
  "role": "user",
  "content": [
    {"type": "text", "text": "What is the bug in this screenshot?"},
    {"type": "image_url", "image_url": {"url": "https://example.com/error.png"}}
  ]
}

The URL can also be data:image/png;base64,....

Response

{
  "id": "chatcmpl-...",
  "object": "chat.completion",
  "model": "anthropic/claude-sonnet-5.5",
  "choices": [
    {
      "index": 0,
      "message": {"role": "assistant", "content": "let can be reassigned, const cannot..."},
      "finish_reason": "stop"
    }
  ],
  "usage": {"prompt_tokens": 32, "completion_tokens": 58, "total_tokens": 90}
}

Behaviour notes

• tool_choice without tools is dropped, so the model does not reject the request.
• Null characters (\u0000) in text are removed before forwarding, because some models reject them.
• A very small max_tokens is raised to the minimum the model accepts.