recording-midi-data

midi-recorder records everything that comes in on your MIDI inputs and saves it as a standard MIDI file. There is nothing to press: the first note starts a recording, and when you stop playing the file is closed and saved.

Each file is named after the moment the recording started: 2026-10-05_21-35-12.mid.

• Notes, sustain and other pedals, channel and polyphonic aftertouch, pitch bend and all other messages are kept. Only clock and active_sensing are dropped.
• Listens on all MIDI inputs at once; inputs plugged in later are picked up automatically.
• Plays back in real time in any DAW (1 tick = 1 ms).
• Crash safe: a journal is written while you play, and recovered at the next start.
• Command line, plus a small status window.
• Linux, Windows and macOS.

There is no live display of the notes; open the finished file in your DAW for that.

Install

pipx install midi-recorder        # or: pip install midi-recorder

Python 3.10 - 3.12. The MIDI backend python-rtmidi has no prebuilt wheels for Python 3.13 yet, so on 3.13 pip would have to compile it.

Quick start

midi-recorder --list              # show the MIDI inputs it can see
midi-recorder -o ~/midi           # record into ~/midi, stop with Ctrl-C
midi-recorder-gui -o ~/midi       # the same, with a status window

midi-recorder --list prints a short error instead of a traceback when no MIDI backend is available.

If you leave out -o, files go into the current directory.

How a recording works

1. A recording starts with the first event from any input.
2. It ends when no event has arrived for --idle seconds (default 30), and no note is held and no sustain pedal is down. While a key or the pedal is held, it waits up to --max-hold seconds (default 120) instead, so a long held chord does not cut the file, and a stuck note cannot keep it open forever.
3. The file is written as a type 1 MIDI file: a tempo track plus one track per input, named after the input. With only one keyboard you get one note track.
4. Ctrl-C (or closing the window) saves the recording in progress before exiting.

Timestamps are taken when the event reaches the program, so their accuracy is whatever your operating system's MIDI driver and scheduling give you, typically about a millisecond.

Options

Both programs take the same options.

Status window

midi-recorder-gui shows whether it is recording (with the number of events and the elapsed time) or ready, the inputs it is listening on, the recordings saved so far, and a button that opens the output folder. Closing the window saves the recording in progress.

If it crashes

While you play, every event is also appended to YYYY-MM-DD_HH-MM-SS.journal in the output folder, and the file is flushed as events arrive. When a recording is saved normally the journal is deleted. If the program or the computer dies mid-recording, the next start turns each leftover journal into a .mid file and logs a warning. A write that was cut off by the crash is skipped. If a .mid with that name already exists, the recovered file gets _recovered in its name instead of overwriting it.

Run only one recorder per output folder, otherwise one could mistake the other's live journal for a crash.

Logging

Messages go to stderr and to <outdir>/midi_recorder.log, which is rotated at 1 MB and keeps 3 old files. A failing MIDI backend or an input that cannot be opened is logged once, not every few seconds.

Development

git clone https://github.com/maarten-boot/recording-midi-data
cd recording-midi-data
make check        # ruff, ruff format --check, mypy --strict, pytest (creates .venv)
make all          # the same from a fresh venv, also runs ruff format
make build        # checks, then builds the sdist and wheel into dist/ and runs twine check
make run ARGS="-o ~/midi"

make with no target lists everything. The checks run on Linux, Windows and macOS in GitHub Actions.

mido ships no type information, so a minimal stub for the part this program uses lives in stubs/mido/.

License

MIT, see LICENSE.

Thanks Claude.