milkeyskills 0.1.0

Milkey skills infrastructure SDK for connecting multi-skill AI workflows into OpenAI, Anthropic, Gemini, and other AI stacks.

milkeyskills

Milkey Skills SDK for Python helps teams connect Milkey skills infrastructure into existing AI products and agent workflows.

Use it when you want to:

• plug Milkey skills into existing OpenAI-compatible clients
• add reusable skill execution to Anthropic, Gemini, or other AI stacks
• keep your model provider client while adding Milkey-managed skills as tools

Install

pip install milkeyskills

If your app already uses the OpenAI Python SDK:

pip install milkeyskills openai

Quick Start

import os

from openai import OpenAI

from milkeyskills import milkey

openai_client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

milkey_client = milkey.create_client(
    base_url=os.environ["MILKEY_BASE_URL"],
    api_key=os.environ["MILKEY_API_KEY"],
)

tools = milkey.openai.chat.tools(client=milkey_client)

messages = [
    {
        "role": "user",
        "content": "Find the best Milkey skill for PostgreSQL query optimization.",
    }
]

first = openai_client.chat.completions.create(
    model="gpt-4.1",
    messages=messages,
    tools=tools,
)

assistant = first.choices[0].message
messages.append(
    {
        "role": "assistant",
        "content": assistant.content or "",
        "tool_calls": [
            {
                "id": tool_call.id,
                "type": "function",
                "function": {
                    "name": tool_call.function.name,
                    "arguments": tool_call.function.arguments,
                },
            }
            for tool_call in assistant.tool_calls or []
        ],
    }
)

if assistant.tool_calls:
    tool_messages = milkey.openai.chat.messages(
        {
            "tool_calls": [
                {
                    "id": tool_call.id,
                    "function": {
                        "name": tool_call.function.name,
                        "arguments": tool_call.function.arguments,
                    },
                }
                for tool_call in assistant.tool_calls
            ]
        },
        milkey_client,
    )
    messages.extend(tool_messages)

    second = openai_client.chat.completions.create(
        model="gpt-4.1",
        messages=messages,
        tools=tools,
    )
    print(second.choices[0].message.content or "")
else:
    print(assistant.content or "")

For production use, keep iterating until the model stops returning tool_calls and enforce a max turn count or request timeout. See examples/openai_chat_completions.py for a bounded loop.

Async Client

from milkeyskills import milkey

async with milkey.create_async_client(
    base_url="https://api.milkey.ai",
    api_key="mk_sk_...",
) as client:
    tools = await client.list_tools()

Supported Integrations

• OpenAI-compatible chat completions
• OpenAI responses and realtime-style hosted tool delivery
• Anthropic inline and hosted MCP integrations
• Gemini function-calling integrations, Gemini Interactions hosted MCP helpers
• Framework-agnostic inline tool helpers (milkey.inline_tools)

Provider Matrix

Provider	API Variant	mode="auto"	Supported Modes	Recommended Helper
OpenAI	Chat Completions	inline	inline, hosted	milkey.openai.chat.tools(...)
OpenAI	Responses API	hosted	inline, hosted	milkey.openai.responses.tools(...)
OpenAI	Realtime	hosted	hosted	milkey.openai.realtime.tools(...)
Anthropic	Messages API	hosted	inline, hosted	milkey.anthropic.config(...)
Gemini	generateContent	inline	inline	milkey.gemini.config(...)
Gemini	Interactions API	hosted	hosted	milkey.gemini.interactions.config(...)
Inline	Custom tool loops	inline	inline	milkey.inline_tools.tools(...)

Versioning

This package follows Semantic Versioning.

pip install --upgrade milkeyskills

Development

cd python
pip install -e ".[dev]"
pytest
ruff check src tests
mypy src