API referenceDraft

Chat completions

Generating replies for a conversation, with or without streaming.

Draft

The hosted API is a planned service with a draft specification. The local API described alongside it already exists and runs on your own machine.

POST/v1/chat/completionsRuns one KOLDOS turn for the last user message: routing, context, tools and verification, not just the model.

Request body

Field	Type	Required	Description
`model`	string	Yes	Model ID, for example koldos-7b.
`messages`	array	Yes	Conversation so far. KOLDOS takes the last user message as the request for the turn.
`temperature, max_tokens, top_p`	number	No	Accepted so standard clients keep working, and ignored: KOLDOS sets its own budgets for each turn.
`stream`	boolean	No	Send the reply as server-sent events while it is generated.

Example request body

{
  "model": "koldos-7b",
  "messages": [
    { "role": "user", "content": "What does src/app.py do?" }
  ]
}

Response shape

{
  "id": "chatcmpl_example",
  "object": "chat.completion",
  "created": 1767225600,
  "model": "koldos-7b",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "..." },
      "finish_reason": "stop"
    }
  ],
  "usage": { "prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0 }
}

What a request becomes

KOLDOS reads the last user message as the request for one turn. The turn decides its own route, context and tool budget, so there is no system prompt or sampling setting to tune from the request.

temperature, max_tokens, top_p are accepted so that standard clients keep working, and ignored.

Streaming

Set stream to true to receive server-sent events while the reply is generated. Each event carries a small delta; the stream ends with data: [DONE].

Event stream

data: {"id":"chatcmpl_example","choices":[{"index":0,"delta":{"content":"..."}}]}

data: {"id":"chatcmpl_example","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

data: [DONE]