bt-proxy

ESPHome-compatible Bluetooth Proxy for Raspberry Pi.

This implements the ESPHome Bluetooth Proxy functionality in Python, allowing a Raspberry Pi to act as a BLE proxy for Home Assistant. It speaks the ESPHome Native API protocol so Home Assistant discovers and uses it exactly like an ESP32-based Bluetooth proxy.

WARNING: This project was coded largely with the assistance of an LLM. It works for me, but your mileage may vary.

Features

• BLE scanning — passive and active scan modes, raw advertisement forwarding
• Active connections — GATT connect/disconnect, service discovery, read/write characteristics and descriptors, notifications
• mDNS discovery — automatically advertised so Home Assistant finds it
• ESPHome Native API — wire-compatible with aioesphomeapi / Home Assistant ESPHome integration

Requirements

• Raspberry Pi (or any Linux machine) with a Bluetooth adapter
• uv package manager
• BlueZ (installed by default on Raspberry Pi OS)

Installation

git clone https://github.com/denvera/bt-proxy.git /opt/bt-proxy
cd /opt/bt-proxy
uv sync

Usage

uv run python -m bt_proxy

Options

Flag	Default	Description
--name	bt-proxy	Device name (used in mDNS and API)
--friendly-name	Bluetooth Proxy	Human-readable name
--port	6053	API server TCP port
--max-connections	3	Max concurrent BLE GATT connections
--adapter	system default	Bluetooth adapter (e.g. hci0)
--log-level	INFO	Logging verbosity

Example

uv run python -m bt_proxy --name living-room-proxy --friendly-name "Living Room BT Proxy" --log-level DEBUG

How It Works

1. Starts a BLE scanner using bleak
2. Advertises itself via mDNS as _esphomelib._tcp.local.
3. Listens on TCP port 6053 for ESPHome Native API connections
4. When Home Assistant connects, it forwards BLE advertisements and handles GATT operations

Note: This uses the ESPHome Native API plaintext variant (no encryption). The Noise-encrypted protocol is currently not supported.

Running with Docker

Pre-built images are available for amd64, arm64, and arm/v7 (Raspberry Pi 2+):

docker run -d \
  --name bt-proxy \
  --restart unless-stopped \
  --net=host \
  --privileged \
  -v /var/run/dbus:/var/run/dbus \
  ghcr.io/denvera/bt-proxy

Pass any CLI options after the image name:

docker run -d \
  --name bt-proxy \
  --restart unless-stopped \
  --net=host \
  --privileged \
  -v /var/run/dbus:/var/run/dbus \
  ghcr.io/denvera/bt-proxy \
  --name living-room-proxy --friendly-name "Living Room BT Proxy" --log-level DEBUG

--net=host is required for mDNS discovery. --privileged grants access to the Bluetooth adapter — alternatively use --cap-add=NET_ADMIN --cap-add=NET_RAW with explicit device mounts.

Running as a Service

Copy the unit file and enable it:

sudo cp bt-proxy.service /etc/systemd/system/
sudo systemctl daemon-reload
sudo systemctl enable --now bt-proxy

Check status / logs:

sudo systemctl status bt-proxy
journalctl -u bt-proxy -f

Scanning modes: active vs passive

Home Assistant requests passive scanning by default. bt-proxy honours that when BlueZ supports it, and otherwise falls back to active scanning automatically — so things keep working either way. If passive isn't available you'll see a one-off warning in the log followed by falling back to active scanning; that's harmless.

Passive is generally the better default. It's lighter-weight on the proxy itself (the radio only listens, never transmits, which also reduces RF congestion when several proxies are around), and it can be gentler on some battery sensors:

• Active scanning may send a scan request to a scannable advertiser, prompting it to transmit an extra scan-response packet. Passive never asks for one.
• For non-connectable beacons (they can't be scan-requested), and for devices the proxy stays connected to — once connected, the connection dominates a device's power use, not the scan mode. It mainly matters for advertisement-only sensors that happen to be scannable.

Enabling passive scanning

Passive scanning needs BlueZ's experimental features (and BlueZ ≥ 5.56 with Linux kernel ≥ 5.10, which most current systems already have). Run bluetoothd with --experimental.

Rather than editing the packaged unit, drop in an override so updates don't clobber it. Create /etc/systemd/system/bluetooth.service.d/experimental.conf:

[Service]
ExecStart=
ExecStart=/usr/libexec/bluetooth/bluetoothd --experimental

The empty ExecStart= is required to clear the inherited command before setting the new one. Then reload and restart:

sudo systemctl daemon-reload
sudo systemctl restart bluetooth

Match the bluetoothd path to your system — some distros use /usr/lib/bluetooth/bluetoothd. Check with systemctl cat bluetooth | grep ExecStart.

Once enabled, the proxy logs Starting BLE scanner (configured=passive, mode=passive) and the active-fallback warning goes away.

Reliability on Raspberry Pi

Undervolting or brownout/undervoltage conditions on Pi's produce exactly the symptoms that look like flaky bluetooth, often manifested as HCI command timeouts (dmesg: hci0: Opcode 0x200c failed, tx timeout), Frame reassembly failed, dropped connections, even an unresponsive bluetoothd.

Before blaming bluetooth, check power:

vcgencmd get_throttled      # want 0x0 (or at most 0x50000 = a brief boot-time
                            # dip); a non-zero bit 0 means under-voltage NOW

Use a proper 5.1 V / ≥2.5 A supply and a short, thick cable. Undervoltage under load (0x...1 / 0x...5) is the most common cause of "random" BT failures on a Pi.

The proxy also backs off instead of hammering the adapter when connections keep failing, which both protects a flaky controller and keeps the logs readable.

Adapter powered off / blocked by rfkill (common on a fresh install)

If the proxy logs the Bluetooth adapter is powered off or blocked / No powered Bluetooth adapters found ... POWERED_OFF, the adapter simply isn't powered — it's not a wedged controller. On a fresh Raspberry Pi OS install the onboard Bluetooth is often soft-blocked by rfkill. Unblock it on the host:

sudo rfkill unblock bluetooth
sudo hciconfig hci0 up          # only if it's still down afterwards

This is a host-level fix — even a --privileged container can't power on an rfkill-blocked host adapter. The proxy detects this case and keeps retrying (every 60s), so once you unblock it the scanner starts on its own without a restart.

rfkill may not be installed by default (sudo apt install rfkill). The unblock is saved across reboots (via systemd-rfkill), so it's usually a one-time fix.

When bluetooth appears unresponsive

Two distinct failures can occur on the onboard radio, with different fixes:

• Controller unresponsive — dmesg shows hci0: Opcode 0x200c failed: -110/-16. Restarting bluetoothd is not enough; reset the adapter:

  sudo hciconfig hci0 reset    # or: down && up, or reboot
  

• Daemon unresponsive — the proxy logs that it looks like "bluetoothd is wedged (phantom connection / no D-Bus reply)". Here the daemon needs a restart (an adapter reset alone won't fix it, and neither will restarting bt-proxy):

  sudo systemctl restart bluetooth
  

bt-proxy tries to detect these conditions, log a recovery hint, and back off so it recovers once the underlying issue clears — but it wont reset the adapter or restart bluetoothd itself (those are host-level actions).

Optional: auto-recover an unresponsive bluetoothd (watchdog)

If you want an automatic, albeit somewhat primitive recovery from the unresponsive daemon case, add a small systemd watchdog on the host that restarts bluetooth when the proxy reports it's frozen. For example, a timer that restarts the daemon when the marker shows up in the proxy's logs:

# /etc/systemd/system/bt-proxy-watchdog.service
[Unit]
Description=Restart bluetoothd if unresponsive

[Service]
Type=oneshot
ExecStart=/bin/sh -c 'journalctl -u bt-proxy --since "-3min" | grep -q "bluetoothd is wedged" && systemctl restart bluetooth || true'

# /etc/systemd/system/bt-proxy-watchdog.timer
[Unit]
Description=Periodically check whether bluetoothd needs restarting

[Timer]
OnUnitActiveSec=2min
AccuracySec=30s

[Install]
WantedBy=timers.target

sudo systemctl daemon-reload
sudo systemctl enable --now bt-proxy-watchdog.timer

Adjust the journalctl -u unit name (and add -t/container filters) to match how you run the proxy. Under Docker, point it at the container logs instead, e.g. docker logs --since 3m bt-proxy 2>&1 | grep -q ....

Architecture

bt_proxy/
├── __init__.py        # Package init
├── __main__.py        # Entry point, CLI, mDNS registration
├── proto.py           # Protobuf encoding/decoding, message IDs, wire protocol
├── ble_manager.py     # BLE scanning and GATT connections (bleak)
└── api_server.py      # ESPHome Native API TCP server