Chat completions

POST /v1/chat/completions — parameters, headers, and the response object.

http

POST https://api.webway.example/v1/chat/completions

OpenAI-compatible. An SDK that accepts a base URL needs no other change.

Headers

Header	Required	Value
`Authorization`	yes	`Bearer sk-ww-…`
`Content-Type`	yes	`application/json`
`Webway-ZDR`	no	`required` to refuse retaining providers

Body

Field	Type	Notes
`model`	string	A catalogue name
`messages`	array	`{role, content}`, roles `system` / `user` / `assistant`
`max_tokens`	integer	Caps the completion, and its cost
`temperature`	number	Passed through where the provider supports it
`top_p`	number	Passed through
`stream`	boolean	Server-sent events — see Streaming
`stop`	string or array	Up to four stop sequences
`thinking`	object	`{"type": "enabled" \| "disabled"}` on reasoning models

Unrecognised fields are forwarded to the provider rather than rejected, so a parameter we have not documented still reaches a model that understands it.

Response

json

{
  "id": "chatcmpl-9f2a…",
  "object": "chat.completion",
  "created": 1787526536,
  "model": "anthropic/claude-sonnet-5",
  "choices": [
    {
      "index": 0,
      "finish_reason": "stop",
      "message": {"role": "assistant", "content": "…"}
    }
  ],
  "usage": {
    "prompt_tokens": 41,
    "completion_tokens": 12,
    "total_tokens": 53
  }
}

usage is the same count the ledger prices, so a client that sums it will agree with the Usage screen.

Streamed response

With stream: true, chunks arrive as data: lines holding chat.completion.chunk objects, ending with data: [DONE].