Chat completions

POST /v1/chat/completions

Request body

top_k, min_p and repetition_penalty are not part of the OpenAI API, so the OpenAI SDKs have no argument for them. Pass them in extra_body (Python) or as extra fields in the request body (JavaScript, curl):

response = client.chat.completions.create(
    model="gemma",
    messages=[...],
    frequency_penalty=0.3,
    extra_body={"top_k": 30, "repetition_penalty": 1.05},
)

Keep repetition penalties low for reports and summaries: names and technical terms have to recur, and high values make the model reach for synonyms.

Message roles

Each message in the array has a role that determines how it’s treated.

Example request

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://router.appelon.ai/v1",
    api_key=os.environ["APPELON_API_KEY"]
)

response = client.chat.completions.create(
    model="gemma",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "What is the capital of the Netherlands?"}
    ]
)

print(response.choices[0].message.content)
# → "The capital of the Netherlands is Amsterdam."

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://router.appelon.ai/v1',
  apiKey: process.env.APPELON_API_KEY,
});

const response = await client.chat.completions.create({
  model: 'gemma',
  messages: [
    { role: 'system', content: 'You are a helpful assistant.' },
    { role: 'user', content: 'What is the capital of the Netherlands?' },
  ],
});

console.log(response.choices[0].message.content);
// → "The capital of the Netherlands is Amsterdam."

Response

Returns a completion object with the generated message.

{
  "id": "chatcmpl-abc123",
  "object": "chat.completion",
  "model": "RedHatAI/gemma-4-31B-it-FP8-block",
  "choices": [{
    "index": 0,
    "message": {
      "role": "assistant",
      "content": "The capital of the Netherlands is Amsterdam."
    },
    "finish_reason": "stop"
  }],
  "usage": {
    "prompt_tokens": 28,
    "completion_tokens": 12,
    "total_tokens": 40
  }
}

Streaming

Set stream: true to receive tokens as they’re generated.

stream = client.chat.completions.create(
    model="gemma",
    messages=[{"role": "user", "content": "Tell me a story"}],
    stream=True
)

for chunk in stream:
    print(chunk.choices[0].delta.content, end="")

const stream = await client.chat.completions.create({
  model: 'gemma',
  messages: [{ role: 'user', content: 'Tell me a story' }],
  stream: true,
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content || '');
}

Tool calling

Both chat models support tool calling, in the standard OpenAI shape: pass tools, and the model may answer with tool_calls instead of content. This is what agent frameworks and the n8n AI Agent node rely on.

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Get the current weather for a city",
        "parameters": {
            "type": "object",
            "properties": {"city": {"type": "string"}},
            "required": ["city"],
        },
    },
}]

response = client.chat.completions.create(
    model="gemma",
    messages=[{"role": "user", "content": "Wat voor weer is het in Groningen?"}],
    tools=tools,
)

call = response.choices[0].message.tool_calls[0]
print(call.function.name)       # get_weather
print(call.function.arguments)  # {"city": "Groningen"}

Run the tool yourself, then send the result back as a tool message carrying the same tool_call_id, and call the model again with the extended history. That loop is the whole mechanism behind every coding agent.

Notes

• Both models work, but they use different parsers underneath. If a prompt produces reliable tool calls on one and not the other, try the other before rewriting the prompt.
• Parallel tool calls may come back as several entries in one tool_calls array. Answer each one with its own tool message.
• Streaming delivers tool calls in fragments: arguments arrives as a string built up across chunks, so accumulate before parsing it as JSON.

Tools built on this

• Hermes Agent, a coding agent
• n8n, workflow automation
• Migrating from OpenAI, for anything else that takes a custom base URL