Skip to content

deepseek.md

Latest commit

 

History

History
145 lines (101 loc) · 3.72 KB

deepseek.md

File metadata and controls

145 lines (101 loc) · 3.72 KB

Integration Guide: DeepSeek

Skillware supports DeepSeek chat models via SkillLoader.to_deepseek_tool(). Use the official openai Python SDK pointed at the DeepSeek API base URL.

This adapter is separate from to_openai_tool() even though the wire format is similar. Do not substitute one for the other in application code.

Set DEEPSEEK_API_KEY for the agent loop. Skills that call other APIs during execute() may need additional variables; see API keys for skills and the skill's catalog page.

Quick snippet

import os

from openai import OpenAI

from skillware.core.env import load_env_file
from skillware.core.loader import SkillLoader

load_env_file()

bundle = SkillLoader.load_skill("finance/wallet_screening")
tool = SkillLoader.to_deepseek_tool(bundle)

client = OpenAI(
    api_key=os.environ.get("DEEPSEEK_API_KEY"),
    base_url="https://api.deepseek.com",
)
# Pass bundle["instructions"] as the system message when you start the chat.

How it works

1. Schema adaptation

to_deepseek_tool() returns the same structural tool object used by OpenAI-compatible APIs:

{
  "type": "function",
  "function": {
    "name": "...",
    "description": "...",
    "parameters": { }
  }
}

Parameters come directly from the skill manifest JSON Schema.

2. Function name sanitization

DeepSeek applies the same function-name rules as other OpenAI-compatible tool APIs. Manifest IDs with slashes are normalized (for example compliance/tos_evaluator becomes compliance_tos_evaluator).

Match tool_call.function.name to tool["function"]["name"] from to_deepseek_tool(), not the raw manifest name string.

3. System message (the Mind)

Pass bundle["instructions"] as the system role content so the model knows when to call the skill.

4. Tool calling loop

Use client.chat.completions.create with tools=[deepseek_tool], handle tool_calls, run skill.execute(...), append tool messages, and repeat.

See examples/deepseek_tos_evaluator.py.

Complete example (manual loop)

import json
import os

from openai import OpenAI

from skillware.core.env import load_env_file
from skillware.core.loader import SkillLoader

load_env_file()

bundle = SkillLoader.load_skill("compliance/tos_evaluator")
skill = bundle["module"].TOSEvaluatorSkill()

deepseek_tool = SkillLoader.to_deepseek_tool(bundle)
tool_name = deepseek_tool["function"]["name"]

client = OpenAI(
    api_key=os.environ.get("DEEPSEEK_API_KEY"),
    base_url="https://api.deepseek.com",
)
messages = [
    {"role": "system", "content": bundle["instructions"]},
    {
        "role": "user",
        "content": "Check whether automated crawling is allowed for https://example.com/docs.",
    },
]

response = client.chat.completions.create(
    model="deepseek-chat",
    messages=messages,
    tools=[deepseek_tool],
)

while response.choices[0].message.tool_calls:
    assistant_message = response.choices[0].message
    tool_call = assistant_message.tool_calls[0]

    if tool_call.function.name != tool_name:
        break

    args = json.loads(tool_call.function.arguments)
    result = skill.execute(args)

    messages.append(assistant_message)
    messages.append(
        {
            "role": "tool",
            "tool_call_id": tool_call.id,
            "content": json.dumps(result),
        }
    )

    response = client.chat.completions.create(
        model="deepseek-chat",
        messages=messages,
        tools=[deepseek_tool],
    )

print(response.choices[0].message.content)

Related documents