xiaoai-bridge 0.2.1

Use XiaoAi speakers as a programmable Python voice entrypoint

xiaoai-bridge

Use XiaoAi speakers as a programmable Python voice entrypoint.

简体中文 · Quickstart · Write a Handler · Commands · Troubleshooting

---

What it does

xiaoai-bridge continuously listens to selected XiaoAi speakers, forwards new user questions to your Python handler, then plays the handler response through the same speaker.

You talk to XiaoAi
  ↓
xiaoai-bridge polls XiaoAi conversation records
  ↓
Your handler(question, speaker) runs in your own project
  ↓
xiaoai-bridge plays text or audio through the matching speaker

Use it to:

• Connect XiaoAi speakers to your own LLM or automation logic.
• Write home voice automations in plain Python.
• Share one handler across multiple speakers while still knowing which speaker triggered the request.
• Return text, remote audio URLs, local audio paths, or streaming text chunks.

[!NOTE] This project uses Xiaomi MiNA APIs inspired by idootop/migpt-next. These APIs are not public stable APIs and may be affected by account security policy, device model, firmware version, region, or upstream changes.

xiaoai-bridge appends TTS/audio playback after observing conversation records. It may not intercept or replace XiaoAi's original response. On some devices or scenarios, users may hear XiaoAi's native response first, then the handler response.

Install

With uv

Create your own bot project and install xiaoai-bridge as a dependency:

mkdir my-xiaoai-bot
cd my-xiaoai-bot
uv init
uv add xiaoai-bridge
xiaoai-init

With pip

python -m venv .venv
source .venv/bin/activate
pip install xiaoai-bridge

[!TIP] Keep your handler.py, .env, and any private automation code in your own project. Do not edit site-packages/xiaoai_bridge/handler.py or the package source.

Quickstart

xiaoai-init creates .env, handler.py, and a small .gitignore in the current directory. Existing files are skipped unless you pass --force.

Recommended login method on machines with a desktop browser:

xiaoai-login

This opens Xiaomi login in your browser, waits for you to sign in, then tries to copy userId and passToken from local browser cookies into .env.

If you are on a headless machine or cookie reading is unavailable, manually fill these values in .env:

MI_XIAOMI_USER_ID="your Xiaomi userId"
MI_XIAOMI_PASS_TOKEN="your passToken, including the V1: prefix"

Create or edit handler.py next to .env:

def handler(question: str, speaker):
    print(f"Question: {question}, speaker: {speaker.display_name}", flush=True)
    return f"{speaker.display_name} heard: {question}"

Check login:

xiaoai-check-login

Select the XiaoAi speakers to listen to:

xiaoai-select

Start the bridge:

xiaoai-bridge

You can also override the handler from the command line:

xiaoai-bridge --handler ./handler.py:handler
xiaoai-bridge --handler my_bot.handlers:handler

Handler priority:

CLI --handler > MI_HANDLER > built-in demo handler

Get passToken

Recommended: browser login command

On a desktop machine, run:

xiaoai-login

The command opens https://account.xiaomi.com/, waits while you sign in, then reads Xiaomi cookies from local browsers using browser-cookie3. If successful, it updates .env automatically:

MI_XIAOMI_USER_ID="..."
MI_XIAOMI_PASS_TOKEN="V1:..."

If this fails because the machine is headless, the browser is unsupported, cookies are encrypted, or no desktop browser is available, use the manual path below.

Manual copy

1. Open https://account.xiaomi.com/ in a browser and sign in.
2. Open Developer Tools.
3. Find Cookies / Storage for https://account.xiaomi.com.
4. Copy:
   • userId
   • passToken
5. Write them to .env:

MI_XIAOMI_USER_ID="..."
MI_XIAOMI_PASS_TOKEN="V1:..."

If Chrome does not show passToken, try Firefox. Copy the V1: prefix as part of passToken.

Select XiaoAi speakers

Run the interactive selector:

xiaoai-select

Keys:

Key	Action
↑ / ↓	Move cursor
Space	Select / deselect
a	Select all / none
Enter	Save to .env
q	Cancel

The selector updates:

MI_SPEAKER_SN="sn1,sn2,..."
MI_SPEAKER_MAC="mac1,mac2,..."

Write a Handler

Create a handler in your own project, for example handler.py:

from xiaoai_bridge.mina_client import MiNADevice


def handler(question: str, speaker: MiNADevice) -> str | None:
    print(f"Question: {question}, speaker: {speaker.display_name}", flush=True)
    return f"{speaker.display_name}, you asked: {question}"

Configure it with one of these forms:

MI_HANDLER="./handler.py:handler"
MI_HANDLER="/absolute/path/to/handler.py:handler"
MI_HANDLER="my_bot.handlers:handler"

If the callable name is omitted, handler is used by default:

MI_HANDLER="./handler.py"

speaker commonly contains:

Field	Meaning
speaker.display_name	Speaker name, preferring alias/name
speaker.serial_number	SN
speaker.mac	MAC address
speaker.hardware	Hardware model, for example LX06
speaker.device_id	MiNA device id
speaker.miot_did	Mi Home did

Branch by speaker

def handler(question: str, speaker) -> str | None:
    if speaker.display_name == "Living Room XiaoAi":
        return "This reply is from the living room speaker."
    return f"{speaker.display_name} received it."

Async handler

async def handler(question: str, speaker) -> str | None:
    answer = await your_llm_call(question)
    return answer

Streaming handler

import asyncio


async def handler(question: str, speaker):
    async for chunk in ask_your_llm_stream(question):
        yield chunk


async def ask_your_llm_stream(question: str):
    for chunk in ["First sentence.", "Second sentence.", "Third sentence."]:
        await asyncio.sleep(0.5)
        yield chunk

Each non-empty yielded chunk triggers one XiaoAi TTS playback.

[!TIP] Yield sentences or short paragraphs, not tokens or single characters. XiaoAi TTS is not a WebSocket audio stream; tiny chunks cause frequent short playback segments.

Return a remote audio URL

def handler(question: str, speaker) -> str | None:
    return "https://example.com/reply.mp3"

Return a local audio path

def handler(question: str, speaker) -> str | None:
    return "/Users/example/Music/reply.mp3"

XiaoAi speakers cannot read files directly from your computer. xiaoai-bridge starts a lightweight HTTP server and maps the file to:

http://<your-lan-ip>:8765/audio/<token>.mp3

If the speaker cannot access that address, set:

MI_PUBLIC_BASE_URL="http://your-reachable-host:8765"

Or return an audio URL that is already reachable by the speaker.

Commands

Command	Purpose
xiaoai-init	Create .env, handler.py, and .gitignore in a user project
xiaoai-login	Open Xiaomi login in a browser and write userId/passToken to .env
xiaoai-bridge	Start the bridge and listen to selected speakers
xiaoai-bridge --handler ./handler.py:handler	Start with a specific handler
xiaoai-select	Interactively select one or more XiaoAi speakers
xiaoai-check-login	Check Xiaomi login, device list, and selected speakers
xiaoai-test-speak	Play default test TTS
xiaoai-test-speak "hello"	Play custom test TTS

Source checkout commands

If you are developing this repository itself, use uv run:

uv sync --dev
uv run xiaoai-bridge --handler ./handler.py:handler
uv run ruff check .
uv run pytest

Token expiration

Normally, if cached serviceToken expires, the program tries to refresh it with passToken from .env.

If passToken is also invalid, you may see:

• XiaoAi no longer plays your configured response.
• Console errors such as 401, XiaomiAuthError, or login failed.
• xiaoai-check-login fails.

Recovery:

# 1. Get a fresh userId / passToken from browser cookies and update .env

# 2. Delete old serviceToken cache
rm -f .data/token_cache.json

# 3. Check login
xiaoai-check-login

# 4. Restart the bridge
xiaoai-bridge

Runtime behavior

On startup, the program:

1. Reads .env.
2. Loads the configured handler from MI_HANDLER or --handler.
3. Uses Xiaomi login state to obtain a MiNA serviceToken.
4. Lists devices and matches selected speakers.
5. Initializes conversation cursors without replaying old records.
6. Polls new conversations every MI_POLL_INTERVAL_SECONDS.
7. Calls handler(question, speaker) for each new question.
8. Plays TTS or audio based on the handler result.

Troubleshooting

No sound

Check login first:

xiaoai-check-login

Then test TTS:

xiaoai-test-speak "test sound"

If the command succeeds but there is no sound, check:

• The selected speaker is the one you are testing.
• The speaker is online and not in an abnormal playback state.
• The speaker volume is not zero.

Handler cannot be loaded

Check MI_HANDLER:

MI_HANDLER="./handler.py:handler"

Make sure:

• the file exists relative to the directory where you run xiaoai-bridge;
• the callable exists and is named handler or explicitly named after :;
• for module:callable, the module is importable in the current environment.

No user questions printed

Confirm:

1. xiaoai-select selected the correct device.
2. You asked the selected speaker a question that produces a normal answer, not only the wake word.
3. The bridge is running:

xiaoai-bridge

Login failed

Prefer passToken login. Frequent automatic account/password login may trigger Xiaomi risk control.

If login fails:

rm -f .data/token_cache.json
xiaoai-check-login

If it still fails, refresh MI_XIAOMI_USER_ID and MI_XIAOMI_PASS_TOKEN.

Current boundaries

• Only MiNA-related capabilities are implemented; full MIoT RC4 protocol support is not included.
• Only new questions after startup are processed; old records are not replayed.
• Streaming text is segmented TTS playback, not true audio streaming.
• Local audio playback depends on network reachability. If the speaker cannot access the generated URL, set MI_PUBLIC_BASE_URL or return a remote URL.
• Xiaomi APIs may change with time, region, account security policy, or device firmware.

Security and privacy

• Do not commit .env, .data/token_cache.json, passToken, or serviceToken to a public repository.
• passToken is a login credential. Refresh it if it expires or leaks.
• Keep handler.py private if it contains personal automation logic, keys, or local service URLs.