Deploy MeiMei Phone Assistant with a real phone, a live backend, and a clear setup path.

Setup, pairing, test calls, firmware updates and troubleshooting for the USB model and the Wi-Fi model.

How MeiMei Phone Assistant handles a call

The phone stays on the real mobile network. MeiMei Phone Assistant pairs over Bluetooth HFP, streams call audio and state to your self-hosted backend over USB or Wi-Fi, and lets the backend agent speak back through the handset.

What happens after payment

Stripe confirms your payment and the order appears on your Account page. We then prepare your devices, ship them within the lead time of your package and update the order status as it moves. Setup questions, the legal guarantee and returns are handled by e-mail through the Support page.

How does MeiMei Phone Assistant connect to my phone?

The device pairs to the phone over Bluetooth HFP, like a hands-free kit. The phone stays on the real carrier network while the device forwards call state and audio to your self-hosted backend over a USB cable (USB model) or over your Wi-Fi (Wi-Fi model).

What does first-time setup involve?

Download the MeiMei appliance bundle from the Support page and run its setup wizard (setup.sh on macOS or Linux, setup.ps1 on Windows) with Docker running. It asks for your AI provider keys and a control-panel password, offers a few optional features, then starts the assistant. Next, connect the device: a USB model stays on its cable and uses the small USB relay program the wizard tells you to start; a Wi-Fi model gets your Wi-Fi details once from the MeiMei Firmware Updater. Finally pair your phone over Bluetooth and place a test call.

Which AI providers can I use?

For the conversation, bring your own key for OpenAI, Anthropic Claude, Google Gemini, xAI Grok, Mistral, Meta Muse Spark (through OpenRouter) or OpenRouter, run a local model with Ollama, or use any OpenAI-compatible endpoint. Listening needs a speech-recognition key from OpenAI, Meta or Deepgram, or a whisper-compatible server that you run yourself; this release has no built-in offline speech recognition. Spoken replies use the built-in online Edge voice with no key, or a voice from a provider you configure. A single OpenAI key covers all three stages.

What does it cost to run?

The device is a one-time purchase and we charge nothing per call. Each AI provider bills you directly under your own account; with MeiMei AI, MeiMei bills your usage per use instead. Examples, prices as published by the providers in September 2026 and billed to you by the provider: GPT-Live-1 realtime voice $0.05 per minute; Muse Spark 1.3 $1.25 per million input tokens and $4.25 per million output tokens; Muse Voice Transcribe speech recognition $0.18 per hour of audio (Romanian is not supported). Prices can change; check the provider before you choose. The Models tab of the control panel shows an estimate per device.

What are "Muse Live" and realtime voice, and which should I use for Romanian?

By default every call uses the speech pipeline: speech recognition, then your conversation model, then speech synthesis. "Muse Live" in MeiMei is that pipeline built from Meta services (Muse Voice Transcribe, then Muse Spark, then text-to-speech); it is not a Meta voice-to-voice service, because Meta offers no such developer API. Muse contributor-tier models are for testing only and are not offered in customer installs. Realtime voice is a separate, optional mode in preview: per device, in the Models tab of the control panel, GPT-Live-1 with your own OpenAI key answers the call directly, with EU processing available; if a realtime session fails, the pipeline takes over. Romanian is not yet confirmed for GPT-Live-1, so keep the pipeline for Romanian callers and switch realtime on only after your own test call.

Is there a subscription or hosted service?

No. The device is a one-time purchase and the assistant software runs entirely on your own machine. Your AI provider keys stay in your local configuration, there is no telemetry, and the only recurring cost is whatever your chosen AI provider charges for usage (or, if you choose MeiMei AI, its pay-per-use price). Firmware and software updates are free.

Can the agent place outgoing calls too?

Yes. When a device session is connected, the backend can send a dial command to the paired phone — including through the MCP endpoint, so Claude or any MCP client can place calls and send SMS with per-device tokens. Incoming and outgoing calls use the same audio pipeline and agent routing.

Where do call history and transcripts live?

In your own self-hosted backend, not on the device and not in any cloud service. The backend keeps transcripts, summaries, caller identity, and per-caller agent memory, and you manage them from the web control panel on your machine.

How are updates and recovery handled?

Firmware updates are free and are installed only over the USB cable with the MeiMei Firmware Updater desktop app (version 0.2.0 or later, for macOS and Windows), never over Wi-Fi or the internet: the app downloads the update made for your unit and writes it through the cable. There is no call limit: the device does not count calls and never requires an update after a number of calls, so you decide when to update. On the Wi-Fi model, connect the cable to the main board, the one with the screen. See section 7 of the manual below.

The deployment manual is available in every supported website language.

This manual covers a complete first deployment: install the assistant software on your computer, connect the device, pair your phone, make a test call, tune the assistant in the control panel, and keep the device updated.

1. Install the assistant software

The assistant runs in Docker on a computer you own that stays on (Windows, macOS or Linux). The MeiMei appliance bundle contains a setup wizard that does the configuration for you.

  1. Install Docker Desktop (Windows, macOS) or Docker Engine with Compose v2 (Linux), start it, and wait until it reports that it is running.
  2. Download the MeiMei appliance bundle from the Support page, unpack it, and open a terminal in its folder.
  3. Run ./setup.sh (macOS, Linux) or .\setup.ps1 (Windows PowerShell). The wizard asks for your conversation provider and key, a speech-recognition key, optional features and a control-panel password, then starts the assistant with docker compose up -d.
  4. Open the control panel at http://localhost:8100/phone-assistant-v2/ and sign in as admin with the password you chose. Later, docker compose down and docker compose up -d in the same folder stop and start the assistant; run the wizard again to change providers or keys.

2. Power and place the device

There are two models. The USB model is one board with the screen; it connects to your computer by cable and has no Wi-Fi. The Wi-Fi model has the same main board with the screen plus a second Bluetooth module, and joins your Wi-Fi. Place either within reliable Bluetooth range of the phone it will answer for.

  1. Connect the device with the supplied USB data cable; on the Wi-Fi model, use the USB port of the main board, the one with the screen. Charge-only cables do not work.
  2. Wait for the start-up screen: it shows MEIMEI, PHONE ASSISTANT and a 12-character code. That code is your device ID; the label prints it after MM-.
  3. Check the second line of the screen: LINK:USB-SERIAL means a USB model, which talks to your computer through the cable; WIFI:... means a Wi-Fi model, which talks to your computer over your Wi-Fi.

3. Connect the device to the assistant

A USB model reaches the assistant through its cable; a Wi-Fi model reaches it over your Wi-Fi after a one-time setup in the MeiMei Firmware Updater.

  1. USB model: keep it connected to the computer that runs the assistant and start the small USB relay program as the setup wizard shows at the end (Docker Desktop cannot pass USB devices into containers). Only one program can use the device USB port at a time.
  2. Wi-Fi model: open the MeiMei Firmware Updater, press Scan for devices, install an update if it offers one, then use its Wi-Fi step to enter your Wi-Fi name and password. The device needs a 2.4 GHz network; the assistant address is filled in for you. Alternatively, join the setup network shown on the device screen with the password shown under it and open http://192.168.4.1.
  3. When the link works, the device screen shows BACKEND:CONNECTED and the device appears on the Devices page of the control panel.

4. Pair the phone and make a test call

The device pairs with the phone like a Bluetooth hands-free kit. The phone keeps its SIM and stays on the real mobile network.

  1. A new device is ready to pair as soon as it is on. To pair later, hold the device button for 3 to 8 seconds, then release it; pairing stays open for 90 seconds. This works on both models. Holding the button longer than 8 seconds erases the Wi-Fi settings, the assistant address and all phone pairings: the screen shows ERASING..., then ERASED, BT CLEARED, and the device restarts. A short press answers a ringing call (when the assistant is connected) or hangs up, also on both models.
  2. On the phone, choose MeiMei Phone Assistant in the Bluetooth list, confirm the pairing, and allow access to contacts and messages.
  3. From another phone, call the paired phone. The assistant answers (with the AI disclosure turned on, it first says that it is an AI assistant), and the call and its transcript appear in the control panel.

5. Tune the assistant in the control panel

Each device has its own workspace in the control panel, so every phone line can run its own agent, voice, models and answering policy.

  1. On the General tab, enable the assistant, choose the auto-answer policy and preferred language, and set the greeting line.
  2. On the Agent, STT and TTS tabs, pick the agent and the speech provider, model and voice for this line, then use "Test provider" to confirm the credentials work.
  3. On the Models tab, choose the conversation model and, if you want it, switch on realtime voice (preview, off by default); the tab also shows a cost estimate for this device.
  4. Set Working Hours if the assistant should only answer at certain times, and manage contacts, call logs, SMS auto-reply and allow/block lists from the top navigation.

6. Keep your details ready for support

Support is by e-mail. A few details let us help you on the first reply.

  1. Keep your order ID (shown on your Account page) and the device ID (MM- plus 12 characters, on the label and in the MeiMei Firmware Updater).
  2. Note the phone model and its operating-system version, and what the device screen showed when the problem happened.
  3. Go through the troubleshooting list below first; if that does not help, write to support from the Support page.

7. Firmware updates

Firmware updates are free. They are installed only over the USB cable with the MeiMei Firmware Updater desktop app, version 0.2.0 or later, never over Wi-Fi or the internet. The device checks that every update is genuine before it installs it.

  1. Install the MeiMei Firmware Updater (macOS or Windows) only from the Support page. If your computer warns that the app is damaged or from an unidentified developer, do not bypass the warning; contact support.
  2. Connect the device with its USB data cable (Wi-Fi model: to the main board, the one with the screen). For a USB model, stop the USB relay program first. Press Scan for devices, then Install update, and keep the cable connected until the app shows the result. Never unplug the device or let the computer sleep during an update; if an update is interrupted, the device waits in RECOVERY MODE and you simply run the update again.
  3. Good to know: there is no call limit. The device does not count calls and never requires an update after a number of calls; you decide when to update. If the screen shows UPDATE REQUIRED, the device could not open its protected settings storage and restarts into RECOVERY MODE: install the update, and contact support if it happens again.
  4. The updater needs an internet connection on your computer to download the update made for your device. Do not use other flashing tools: a secured device refuses them.

Troubleshooting checkpoints

  • The device never appears in the control panel: use a USB data cable, not a charge-only one. USB model: make sure exactly one USB relay program is running. Wi-Fi model: run the Wi-Fi step of the MeiMei Firmware Updater again (2.4 GHz networks only).
  • The screen shows BACKEND:CONNECTING or RETRY: check that the computer is on, Docker is running and the assistant is started (docker compose ps in the appliance folder), then check the USB relay (USB model) or the Wi-Fi settings (Wi-Fi model).
  • Phone pairing fails: remove old MeiMei Phone Assistant entries from the phone Bluetooth list, then hold the device button for 3 to 8 seconds and release it to open pairing again (both models).
  • Calls connect but the assistant does not hear the caller: a speech-recognition key is missing or wrong. Run the setup wizard again, or use "Test provider" on the STT and TTS tabs.
  • The screen shows UPDATE REQUIRED or RECOVERY MODE: install the update with the MeiMei Firmware Updater (section 7).
  • Replies feel wrong or generic: review the agent selection and the per-device prompts on the Agent tab.

When to contact support

If the checkpoints above do not solve it, write to support by e-mail (the address is on the Support page) with your order ID, the device ID, your phone model, the step that failed and what the device screen showed.

Support and warranty