Tool calling / function calling with Velqa — OpenAI style

Tool calling lets a model call your functions: you describe tools in the tools parameter, the model replies with tool_calls, you run the matching function, then send the result back in a role: "tool" message. Velqa follows the OpenAI format exactly — just point your base_url at https://api.velqa.dev/v1.

The 4-step cycle

You send the request with the tools list (function schemas).
The model replies with finish_reason: "tool_calls" and one or more tool_calls.
You run the function(s) in your code.
You send the result back in a {"role": "tool", ...} message, then call the model again to get the final answer.

Full Python example (`get_weather` tool)

import json
from openai import OpenAI

client = OpenAI(base_url="https://api.velqa.dev/v1", api_key="sk-...")

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

# Your real implementation would call a weather API.
def get_weather(city):
    return {"city": city, "temp_c": 24, "condition": "sunny"}

messages = [{"role": "user", "content": "What's the weather in Rabat?"}]

# Steps 1-2: the model decides to call the tool
resp = client.chat.completions.create(
    model="kimi-k2.6",
    messages=messages,
    tools=tools,
)
msg = resp.choices[0].message
messages.append(msg)  # keep the tool call in the history

# Step 3: run each tool_call
for call in msg.tool_calls:
    args = json.loads(call.function.arguments)
    result = get_weather(**args)
    messages.append({
        "role": "tool",
        "tool_call_id": call.id,
        "content": json.dumps(result),
    })

# Step 4: the model produces the final answer from the result
final = client.chat.completions.create(
    model="kimi-k2.6",
    messages=messages,
    tools=tools,
)
print(final.choices[0].message.content)

Shape of a `tool_call`

{
  "id": "call_abc123",
  "type": "function",
  "function": {
    "name": "get_weather",
    "arguments": "{\"city\": \"Rabat\"}"
  }
}

The arguments field is a JSON string — remember to parse it (json.loads) before use.

Best practices

Always add the assistant message containing the tool_calls to the history before the tool messages, otherwise the model loses the link.
The tool_call_id of the tool message must match the id of the tool_call.
A model can request multiple tools in parallel: loop over all tool_calls.

Recommended models for tools

Tool calling is especially reliable with kimi-k2.6, minimax-m3 and glm-4.7. See the models list.