Xiaomi MiMo Provider

Xiaomi MiMo provides an OpenAI-compatible API with built-in thinking/reasoning support. Features adaptive thinking mode management.

Install

bashpip install "mtpx[xiaomi]"

This installs the openai SDK:

bashpip install openai

API Key Setup

Option 1: .env file (recommended)

1. Install dotenv support:

bash   pip install python-dotenv

Or with the MTP extra:

bash   pip install "mtpx[dotenv]"

1. Create a .env file in your project root:

text   MIMO_API_KEY=your_key_here

1. Load it in your code before creating the provider:

python   from mtp import Agent

   Agent.load_dotenv_if_available()  # reads .env file

Option 2: System environment variable

bash# Linux/macOS
export MIMO_API_KEY="..."

# Windows PowerShell
$env:MIMO_API_KEY="..."

Quick Start

pythonfrom mtp import Agent
from mtp.providers import Xiaomi

Agent.load_dotenv_if_available()  # loads MIMO_API_KEY from .env

provider = Xiaomi(model="mimo-v2.5-pro")
tools = Agent.ToolRegistry()
agent = Agent(provider=provider, tools=tools)

reply = agent.run_loop("What is 25 * 4 + 10?")
print(reply)

Parameters

Capabilities

Thinking Modes

Xiaomi MiMo has adaptive thinking that automatically manages when reasoning is active:

• `"adaptive"` (default): Thinking is enabled for initial planning, disabled after tool rounds. Finalization uses final_thinking_mode.
• `"enabled"`: Thinking is always on.
• `"disabled"`: Thinking is always off.

python# Adaptive thinking (recommended)
provider = Xiaomi(
    model="mimo-v2.5-pro",
    thinking_mode="adaptive",
    final_thinking_mode="enabled",
)

# Always thinking
provider = Xiaomi(
    model="mimo-v2.5-pro",
    thinking_mode="enabled",
)

# No thinking
provider = Xiaomi(
    model="mimo-v2.5-pro",
    thinking_mode="disabled",
)

Streaming

Xiaomi supports both planning and finalization streaming with reasoning chunks:

pythonfor event in agent.run_loop_events(
    "Calculate 25 * 4",
    max_rounds=2,
    stream_final=True,
):
    if isinstance(event, dict):
        if event.get("type") == "reasoning_chunk":
            print("[thinking]", event["chunk"])
        elif event.get("type") == "text_chunk":
            print(event["chunk"], end="", flush=True)

Full Example

pythonfrom mtp import Agent
from mtp.providers import Xiaomi

Agent.load_dotenv_if_available()

provider = Xiaomi(
    model="mimo-v2.5-pro",
    temperature=0.0,
    thinking_mode="adaptive",
    final_thinking_mode="enabled",
)

tools = Agent.ToolRegistry()
agent = Agent(provider=provider, tools=tools, debug_mode=True)

reply = agent.run_loop(
    "Calculate (25 * 4) + 10",
    max_rounds=3,
)
print(reply)

Notes

• Uses OpenAI-compatible API at https://token-plan-ams.xiaomimimo.com/v1.
• Thinking/reasoning content is captured from reasoning_content field on response messages.
• The adapter automatically disables thinking after tool rounds when in adaptive mode to save tokens.
• Supports extra_body parameter for passing thinking configuration to the API.

Source

src/mtp/providers/xiaomi_provider.py