rdm-card-reader/README.md

# rdm-card-reader

Read 125 kHz RFID cards (RDM6300 reader, EM4100 cards) on a serial port and
print what happens: **card inserted** with the card data, then **card removed**.

Built for a Raspberry Pi workstation, but it runs anywhere with a serial port.

> Starting it on a machine with no reader attached? See **[LAUNCH.md](LAUNCH.md)**
> — it explains the startup sequence, why it exits `1` with no hardware, and how
> to exercise the reader path anyway.

---

## What it does

```
$ rdm-card-reader

-> Opening /dev/ttyAMA0 ...
   OK reader ready on /dev/ttyAMA0 - waiting for cards (Ctrl+C to stop)

----------------------------------------------------------
 CARD INSERTED - 2026-10-01 14:56:54
    id        : 12886197   (0x00C4A0B5)
    type      : 0x0F
    checksum  : 222
    valid     : True
    port      : /dev/ttyAMA0
----------------------------------------------------------
 CARD REMOVED  - 2026-10-01 14:56:54  id: 12886197 (0x00C4A0B5)
```

`id` is the number printed on the card. A card is reported once on insert
(the reader library de-duplicates repeated reads while the card sits on the
antenna) and once on removal.

### Startup sequence

1. **Check dependencies.** If `rdm6300` / `pyserial` cannot be imported, install
them automatically and carry on. A fresh device needs no manual `pip` work.
2. **Find the reader.** Resolve the serial port and open it, identifying the
adapter by serial number or `by-id` rather than trusting `ttyUSB0`.
3. **Start reading.** Read frames in a loop; each valid one becomes a card
event.
4. **Publish it.** In `--serve` mode the event is latched so a web page can
claim it whenever it gets around to asking.

Steps 1 and 2 both fail *loudly with a reason* rather than starting up
half-working, because a reader that silently picks the wrong port or runs
without its library is worse than one that refuses to start.

---

## Install

On a machine **with internet**:

```bash
pip install .
```

`rdm6300` and `pyserial` are pulled in automatically. On Raspberry Pi OS with
an externally-managed Python, add `--break-system-packages` or use a venv:

```bash
python3 -m venv .venv && . .venv/bin/activate && pip install .
```

### On a device with no internet

`pip install .` will **not** work offline. It builds the project, which makes
pip download `setuptools` into an isolated build environment, and that fails
with:

```
ERROR: Could not find a version that satisfies the requirement setuptools>=68
       (from versions: none)
```

The runtime dependencies are bundled in `src/rdm_card_reader/wheels/`, so
install those directly and put `src` on the path — no build step, no network:

```bash
python3 -m venv .venv
.venv/bin/pip install --no-index \
    --find-links src/rdm_card_reader/wheels rdm6300 pyserial
echo "$PWD/src" > "$( .venv/bin/python -c \
    'import sysconfig; print(sysconfig.get_paths()["purelib"])' )/rdm_card_reader_src.pth"
.venv/bin/python -m rdm_card_reader --demo     # verify, no hardware needed
```

`deploy/install.sh` does exactly this, plus the service setup. Running the
app directly also works: it installs anything missing from the same bundled
wheels on first start.

---

## Usage

```bash
rdm-card-reader                     # auto-detect the serial port
rdm-card-reader --port /dev/ttyUSB0 # use a specific port
rdm-card-reader --json              # one JSON object per event
rdm-card-reader --heartbeat 1.5     # slower card-removed detection
rdm-card-reader --list-ports        # show which ports exist, then exit
rdm-card-reader --serve             # serve card data to a web app (see below)
rdm-card-reader --demo              # simulate reads, no hardware needed
```

Also works as a module: `python -m rdm_card_reader`.

| Option | Default | Meaning |
| --- | --- | --- |
| `--port` | auto-detect | Serial port: device path, by-id symlink, serial number or `vid:pid` |
| `--autodetect` | off | Pick the only USB adapter; refuse if there is more than one |
| `--heartbeat` | `0.5` | Seconds without a read before the card counts as removed |
| `--json` | off | Machine-readable output instead of the text block |
| `--list-ports` | – | List candidate and identified ports, then exit |
| `--verbose-ports` | off | With `--list-ports`, show by-id / by-path detail |
| `--demo` | – | Simulated reads; needs no hardware and no `rdm6300` |
| `--demo-count` | `3` | Number of simulated reads in `--demo` |
| `--no-install-deps` | off | Never install missing libraries automatically |
| `--wheel-dir` | bundled | Install from local wheels in this DIR instead of the bundled ones |
| `--version` | – | Print the version |

With `--serve`: `--host` (`127.0.0.1`), `--bridge-port` (`8765`), `--mode`
(`login`), `--tap-ttl` (`10`), `--allow-origin` (repeatable), `--token`,
`--allow-remote`, `--verbose`.

When `--port` is omitted the ports are tried in this order, the same order the
workstation app uses:

```
/dev/ttyS0  ->  /dev/ttyAMA0  ->  /dev/ttyUSB0  ->  /dev/ttyACM0
```

### Output modes

Text mode is for humans and logs. `--json` prints one object per event, which
is what you want when piping into another program:

```json
{"event": "card_inserted", "timestamp": "2026-10-01 14:56:54", "id": 12886197, "hex": "0x00C4A0B5", "type": 15, "checksum": 222, "is_valid": true, "port": "/dev/ttyAMA0"}
{"event": "card_removed", "timestamp": "2026-10-01 14:56:54", "id": 12886197, "hex": "0x00C4A0B5", "type": 15, "checksum": 222, "is_valid": true, "port": "/dev/ttyAMA0"}
```

### Automatic dependency install

The wheels for `rdm6300` and `pyserial` **ship inside this package**
(`rdm_card_reader/wheels/`), so a device with no internet can still set itself
up. On first start the reader installs whatever is missing from those wheels:

```
Reader library not found - checking dependencies...
Missing reader dependencies: rdm6300
Using the wheels bundled with this package:
   /opt/rdm-card-reader/.venv/lib/python3.13/site-packages/rdm_card_reader/wheels
Installing from .../rdm_card_reader/wheels
Dependencies ready: rdm6300
```

No network access is needed. They ship with the package rather than being
fetched at build time, because the build machine and the target device are not
necessarily the same.

Both wheels are **pure Python** (`py3-none-any` / `py2.py3-none-any`), which is
what makes this possible: the same files install on armv7l, aarch64 and x86_64.
This differs from the workstation app's `Files/repository/` cache, which held
architecture-specific binaries (aarch64 manylinux wheels) and so could not be
committed — that cache only ever existed on the Pi devices themselves.

* Nothing is installed when the libraries are already present, so a normal start
  does no network work at all.
* The bundled install runs with `--no-index`, so it works with no route to PyPI
  at all — no proxy or DNS needed.
* Wheels are verified against a bundled `SHA256SUMS` before installation. **A
  wheel that fails its checksum is refused, not installed** — then it falls back
  to PyPI, and if that is unreachable it reports `exit 2` rather than proceeding
  with something unverified.
* Outside a virtualenv pip is given `--break-system-packages`, which Debian and
  Raspberry Pi OS require before they will touch the system Python.
* A failed install does not crash: it reports `exit 2` and tells you the exact
  command to run by hand.

To install from your own wheel directory instead (for example a newer version
than the bundled one), pass `--wheel-dir`. An explicit directory that is
incomplete or fails verification is reported and then PyPI is used, rather than
failing outright. To manage dependencies yourself, pass `--no-install-deps`.

### Exit codes

| Code | Meaning |
| --- | --- |
| `0` | Ran normally, or stopped with Ctrl+C |
| `1` | No serial port could be opened |
| `2` | The `rdm6300` library is missing and could not be installed |
| `3` | The port selector matched more than one adapter (refused to guess) |

---

## Serving card data to a web app (`--serve`)

This is the mode for **authenticating a web page**: the reader runs as a service
on the device, and the page asks it for the card that was just tapped.

```bash
rdm-card-reader --serve --port /dev/rdm-reader --allow-origin https://your.web.app
```

The bridge reports **card ids only**. Deciding which user a card belongs to is
your server's job, not this program's.

### The flow

```mermaid
sequenceDiagram
    participant C as Card
    participant B as rdm-card-reader (Pi)
    participant K as Web page
    participant W as Your server

    C->>B: tap
    B->>B: latch the tap (tap-ttl seconds)
    K->>B: GET /api/card
    B-->>K: {claimable: true, card: {id: ...}}
    K->>B: POST /api/card/claim
    B-->>K: {card: {id: ...}}   (once)
    K->>W: POST /login {card_id}
    W->>W: card_id -> user
    W-->>K: session
```

The point of the *latch* is that the tap stays available for `--tap-ttl`
seconds (default 10), **including after the card is lifted**. A quick tap
therefore cannot be lost because the page happened to poll a moment later.

The point of *claim* is single use. Once a tap is spent, a page reload or a
second tab gets `409 Conflict` instead of a second login, so one physical tap
can never authenticate two sessions.

### Endpoints

| Method | Path | Purpose |
| --- | --- | --- |
| `GET` | `/api/card` | The latched tap: `present`, `claimable`, `consumed`, `age_seconds`, `card` |
| `POST` | `/api/card/claim` | Spend the tap on one login. `200`, or `409` if used/none, `410` if expired |
| `GET` | `/api/events` | Server-Sent Events stream, pushes each event as it happens |
| `GET` | `/api/last` | Most recent card event, or `204` |
| `GET` | `/api/taps` | Recent tap history (`?limit=N`) |
| `GET` | `/api/health` | Reader state, port, uptime, tap counts — for monitoring |
| `POST` | `/api/mode` | `{"mode": "login"}` or `{"mode": "monitor"}` |

`login` mode (default) publishes taps only — a tap is an authentication
attempt. `monitor` mode also publishes removals, which is what you want for
watching a reader rather than logging people in.

### Browser example

```js
const state = await (await fetch("http://127.0.0.1:8765/api/card")).json();
if (state.claimable) {
  const res = await fetch("http://127.0.0.1:8765/api/card/claim", { method: "POST" });
  if (res.ok) {
    const { card } = await res.json();
    await fetch("/login/card", {           // your server decides who card.id is
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ card_id: card.id, hex: card.hex }),
    });
  }
}
```

Or push instead of poll, so a tap logs in immediately:

```js
const events = new EventSource("http://127.0.0.1:8765/api/events");
events.onmessage = (e) => { const tap = JSON.parse(e.data); /* tap.id */ };
```

### Security, and its limits

Read :ref:`the warning below <card-ids-are-not-secrets>`: an EM4100 id is a
*username*, not a password. These controls bound who can *read* the bridge;
they do not make the card itself convincing as a sole credential.

| Control | Behaviour |
| --- | --- |
| Bind address | `127.0.0.1` by default. Anything else needs `--allow-remote`, so an open card-event endpoint cannot appear on the LAN by accident. |
| Origins | Cross-origin browser requests are **refused** unless `--allow-origin` lists that exact origin. Stops another site from reading your taps. |
| Token | `--token` (or `RDM_BRIDGE_TOKEN`) is required from non-loopback callers. |
| Single use | A tap can only ever produce one successful claim. |
| Expiry | A tap older than `--tap-ttl` is refused with `410`. |

Add a second factor — a PIN after the tap — wherever a cloned card would
actually cost you something.

<a name="card-ids-are-not-secrets"></a>
> **Card ids are not secrets.** EM4100 cards broadcast a fixed 40-bit id with no
> challenge-response, and clones are cheap. If holding an id is enough to log
> in, the security is "whoever knows the number". Treat card auth as
> identification, and require something else for anything sensitive.

---

## Running at boot (Raspberry Pi)

```bash
sudo ./deploy/install.sh --origin https://your.web.app
```

That installs a service account, a virtualenv, a systemd unit and a udev rule,
then enables the service. Afterwards:

```bash
systemctl status rdm-card-reader      # state
journalctl -u rdm-card-reader -f      # live log
curl http://127.0.0.1:8765/api/card   # tap a card, then run this
```

### Identifying the right ttyUSB

`/dev/ttyUSB0` is **not** a stable identity — unplug something and the number
changes. Point this at the wrong adapter and you authenticate the wrong reader,
so the tooling is deliberately strict about it.

```bash
rdm-card-reader --list-ports --verbose-ports
```

```
Identified serial ports (best candidate first):
   [score 100] /dev/ttyUSB0  ::  CH340  ::  1a86:7523  ::  S/N AB0XYZ12
       by-id     : /dev/serial/by-id/usb-1a86_USB_Serial-if00-port0
       by-path   : /dev/serial/by-path/pci-0000:01:00.0-usb-0:1.2:1.0-port0
```

Then pin it, most specific first:

```bash
--port /dev/rdm-reader                                              # udev alias
--port /dev/serial/by-id/usb-1a86_USB_Serial-if00-port0            # by device
--port AB0XYZ12                                                     # by serial number
--port 1a86:7523                                                    # by chip
```

`--autodetect` picks the only USB adapter, and **refuses** when there is more
than one — silently choosing between two readers is exactly the failure this
avoids. Verify the wiring in `deploy/99-rdm-reader.rules`.

Exit codes: `0` ok, `1` no port, `2` library missing, `3` port ambiguous.

---

## Status LED

An LED driven by card events, on **GPIO 23** by default — the same pin the
workstation app used.

```bash
rdm-card-reader                    # lit while running, blinks on a bad read
rdm-card-reader --led-pin 17       # a different pin
rdm-card-reader --led-mode presence  # lit only while a card is on the reader
rdm-card-reader --no-led           # no LED at all
```

| Flag | Default | Meaning |
| --- | --- | --- |
| `--no-led` | off | Do not drive a LED |
| `--led-pin` | `23` | GPIO pin (matches the workstation app) |
| `--led-mode` | `error` | See the two readings below |
| `--led-no-error-blink` | off | Do not blink on a bad read |

Two modes, because the light can mean two different things:

| Mode | Behaviour | Use when |
| --- | --- | --- |
| `error` (default) | **Lit whenever the reader is running.** Dark while a card is being read, blinks on a bad read | The light should mean "the reader is alive and ready" |
| `presence` | **Lit only while a card is on the reader** | The light should mean "a card is here" |

Wiring: LED anode to the GPIO pin through a ~330 Ω resistor, cathode to GND.

### It cannot break the reader

Three deliberate choices, because a LED is a convenience and the reader is the
point:

* **Never blocks.** The workstation app toggled the LED directly on the reader
  thread, which stalls card detection. Here every change is queued to a worker
  thread, so a toggle returns immediately even with slow hardware.
* **Never fatal.** On a machine with no GPIO, `gpiozero` raises
  `BadPinFactory`. The app falls back to a no-op LED and says why, rather than
  refusing to run.
* **Never crashes on a dead pin.** A failed toggle is logged and ignored.

Startup reports which of the three states it is in:

```
LED on GPIO 23 (error mode, gpiozero)              # working
LED unavailable: gpiozero not installed            # library missing
LED unavailable: no usable GPIO on pin 23: ...     # no pin factory / pin busy
```

That distinction matters: `BadPinFactory` subclasses `ImportError`, so a naive
check reports "not installed" when the library is present but has no usable pin
provider — which would send you to the wrong fix.

> **If the workstation app also runs on this device, use a different pin** — two
> processes driving GPIO 23 will fight, and the LED will flicker.

`gpiozero` and `colorzero` are installed automatically on first use, from the
bundled wheels, so this works offline. They add ~180 KB. `setuptools`, which
`gpiozero` pulls in as a *packaging* dependency, is deliberately not vendored:
it is never imported at runtime and would add 818 KB for nothing.

---

## Wiring

The RDM6300 talks plain serial at **9600 baud, 8N1**. Connect it to the Pi's
UART (pins 8/10 for TXD/RXD, 5 V and GND) or through a USB-to-serial adapter.

On a Raspberry Pi you usually need to free up the primary UART first:

```bash
sudo raspi-config          # Interface Options -> Serial Port
#   "login shell over serial?"  -> No
#   "serial port hardware?"     -> Yes
sudo usermod -aG dialout $USER   # then log out and back in
```

The GPIO UART appears as `/dev/serial0` (a symlink to `/dev/ttyAMA0` or
`/dev/ttyS0`).

---

## The protocol

Worth knowing if you ever decode the bytes yourself. The reader emits a frame
between `STX` (0x02) and `ETX` (0x03) as **12 ASCII hex nibbles**:

```
0x02  0F 00C4A0B5 DE  0x03
      |  |        |
      |  |        +-- checksum
      |  +----------- card id (4 bytes, big-endian)
      +-------------- type byte
```

Two details that are easy to get wrong:

1. **Field order is type, id, checksum** — the id is *not* first.
2. **The checksum is the XOR of the type byte *and* all four id bytes.** XOR-ing
   only the id bytes produces a mismatch on every card.

```
checksum = type ^ b0 ^ b1 ^ b2 ^ b3
         = 0x0F ^ 0x00 ^ 0xC4 ^ 0xA0 ^ 0xB5
         = 0xDE   (= 222)
```

A frame whose checksum does not match is reported as `INVALID CARD` and is
never treated as a card insert.

There is also a subtle failure mode: a frame with the wrong number of nibbles
is not rejected outright. Two stray nibbles stay in the buffer and corrupt the
*next* read instead of the current one, so a single glitch can look like a
problem with the following card.

---

## Troubleshooting

| Symptom | Fix |
| --- | --- |
| `device not found` on every port | Check the wiring; run `rdm-card-reader --list-ports` |
| `permission denied` | `sudo usermod -aG dialout $USER`, then log out and back in |
| Reader opens but no cards | Card must be EM4100 / 125 kHz; hold it ~2 cm from the coil |
| `INVALID CARD` repeatedly | Noisy supply or long/unshielded wiring; shorten it |
| Every card reported twice | Lower `--heartbeat` if the card is lifted quickly |

---

## Check it works

With no RFID hardware attached, `--demo` exercises the formatting path and
`--list-ports` shows what serial ports exist:

```bash
rdm-card-reader --demo              # simulated reads, no hardware needed
rdm-card-reader --list-ports        # candidate and identified ports
```

To drive the real reader path with no hardware, feed RDM6300 frames into a
pseudo-terminal — see the snippet in **[LAUNCH.md](LAUNCH.md)**.

### Layout

```
src/rdm_card_reader/
    card.py     data model, constants, output formatting (no hardware)
    reader.py   serial port handling, event callbacks
    cli.py      argument parsing and program flow
    demo.py     simulated reads
    ports.py    stable serial port identification
    bridge.py   the HTTP bridge behind --serve
    led.py      status LED
    bootstrap.py  offline dependency install
    wheels/     the bundled wheels (offline install source)
```

### Using it as a library

```python
from rdm_card_reader import open_reader

def on_event(event):
    if event.event == "card_inserted":
        print("card", event.value, event.hex_value)

reader = open_reader(on_event)
if reader is None:
    raise SystemExit("no reader found")

reader.start()   # blocks; call reader.close() to stop
```

---

## License

MIT