Home›Developer Practical

DeepSeek API Function Calling and Structured Output Guide

DeepSeek-V3 natively supports standard OpenAI tool calling schemas and JSON mode. Use these code templates to execute function calls in automated agents.

Maintained by DeepSeekDSH (independent site)Updated: 2026-09-13

On this page

1. Model capability and support

Use deepseek-chat for function and tool execution tasks. DeepSeek-Chat supports native OpenAI-format tool calling schemas. deepseek-reasoner is specialized for pure reasoning and does not support tool calling.

2. Standard 4-step tool calling workflow

Define tool schemas: Provide the tools array containing function names, descriptions, and JSON Schema parameters.
Model decision: The API returns tool_calls with parsed arguments.
Execute locally: Your application invokes the corresponding local function with the provided arguments.
Return result: Append the tool output (with role: "tool") to messages and request the final synthesized answer.

3. Complete executable Python example

import json
from openai import OpenAI

client = OpenAI(
    api_key="sk-your_deepseek_api_key",
    base_url="https://api.deepseek.com"
)

# 1. Local implementation
def get_weather(location: str, unit: str = "celsius"):
    return json.dumps({"location": location, "temperature": 22, "condition": "Sunny", "unit": unit})

# 2. Tool schema
tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Get current weather for a specified location",
        "parameters": {
            "type": "object",
            "properties": {
                "location": {"type": "string", "description": "City name, e.g. London"},
                "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]}
            },
            "required": ["location"]
        }
    }
}]

messages = [{"role": "user", "content": "What is the weather in London today?"}]

# 3. Initial request triggering tool call
response = client.chat.completions.create(
    model="deepseek-chat",
    messages=messages,
    tools=tools
)

message = response.choices[0].message
if message.tool_calls:
    messages.append(message)
    for tool_call in message.tool_calls:
        if tool_call.function.name == "get_weather":
            args = json.loads(tool_call.function.arguments)
            tool_result = get_weather(**args)
            # 4. Append tool execution output
            messages.append({
                "role": "tool",
                "tool_call_id": tool_call.id,
                "content": tool_result
            })

    # 5. Second request producing final answer
    final_response = client.chat.completions.create(
        model="deepseek-chat",
        messages=messages
    )
    print("Final answer:", final_response.choices[0].message.content)

4. Structured JSON Object output mode

When you only require strict JSON formatted output without external tools, set response_format:

response = client.chat.completions.create(
    model="deepseek-chat",
    messages=[
        {"role": "system", "content": "You are a data extraction assistant. Always output valid JSON."},
        {"role": "user", "content": "Extract data: John Doe, 29, Software Engineer in Seattle"}
    ],
    response_format={"type": "json_object"}
)
print(response.choices[0].message.content)

5. Troubleshooting and retry tips

JSON decode error when parsing arguments

Wrap json.loads() in a try...except block. If arguments are malformed, append the error message back to the conversation for auto-repair.

Model fails to trigger tool execution

Ensure the function description clearly matches the user prompt intent, or pass tool_choice="auto" explicitly.