side A, track 2

AP-7

A pocket field recorder for your phone.

An aluminium face, a reel you can grab, a 64×32 dot display and a red record lamp. Press one key and it's recording 48 kHz WAV. Later, turn a tape into text and let brain AI sum it up, tidy it or draw it as a mind map.

In development Coming soon Inspired by the TP-7

Owner's manual

AP-7, all the controls. Point at a line and the recorder shows you where it is.

On the face

  • Start recording. Press again to pause, again to carry on. The red lamp means it's listening.

  • Play or pause the tape that's loaded.

  • Stop. Press it again to rewind to the top.

  • Next or previous tape. The capsule sits at the bottom right of the reel.

  • On the left edge. Hold it up to wind forward, down to rewind, at 8×. You can hear it wind.

  • Touch it to hold the tape, turn it to scrub (clockwise is forward). Touch it while recording and the take pauses: you're off the record until you let go.

  • The M key or the display opens the tape list. Tap a tape to load it and get its keys: share, export, transcribe, open transcript, rename, and delete (tap twice).

Without opening the app

  • Long-press the AP-7 icon on your home screen and pick memo. It's recording before you've finished the thought.

  • Turn it on in the settings tab and an "ap–7 memo" app appears that records at once, even over the lock screen. On a Samsung: Settings › Advanced features › Side button › Double press › Open app › ap–7 memo. (No app gets press-and-hold of the power key, so double press it is.)

In the settings tab

  • The face's finish: silver or graphite. Try both with the switch under the recorder.

  • Off by default. Records through the phone's voice processing with noise suppression: good for speech, may colour music or room sound. Starts with the next take.

  • On by default: the screen stays on while recording or playing. Turn it off to let the screen sleep. The tape keeps rolling either way.

fig. 1 is the recorder, live, right there

AP-7 field recorder − + M

64×32 dot display

red lamp = you're recording

lever: wind or rewind at 8×, you can hear it

grab the reel to scrub. touch it mid-take = off the record

M = tape list

Press ● to record, ▶ to play, ■ to stop. Drag the reel, hold the lever. Works with Tab and Enter too.

side A, track 2, part two

Tape to text

Talk now, read it later.

Long-press a tape and pick transcribe. AP-7 converts it to 16 kHz mono and sends it to the transcription service you chose. The transcript shows up under the tape, with copy text to take it anywhere. It keeps working if you leave AP-7, and a notification tells you when it's ready.

your tape 48 kHz WAV 16 kHz mono only when you tap Gemini Flash
Your 48 kHz tape is converted to 16 kHz mono, uploaded only when you tap transcribe, and comes back as text.
Off by default Uploads only when you tap

Pick your engine

Bring your own API key for the service you pick, or the address of your own server. Press one to see what it needs.

Your keys stay home

API keys are kept on the phone and left out of backups. Cloud services are always reached over https.

Brain AI ✦

One memo, six ways to read it.

Open a transcript and press ✦. Brain AI runs on your own language model server and rewrites the tape: summarize, explain, formalize, casualize, write an article, or draw a mind map you can pan and pinch. Each result streams in as it's written, is kept with the transcript, and can be written again. The tape list marks it ai.

ap7_20261004_190212_saturday rehearsal.wav 01:12

press transcribe

the tape: a band memo, recorded on the bus

✦ brain AI

Pick an action. The answer streams in here.

Sample memo and answers are written for this page so you can see the idea. Your real results depend on your model.

bonus track

Your own server

Rather keep every word at home? Point AP-7 at your own PC. A machine with an NVIDIA GPU, reached over Tailscale, does the transcribing and the brain AI.

tailscale tunnel plain http:// is fine in here, the tunnel is encrypted whisper.cpp :8080 ollama :11434

Then, in AP-7

  1. Transcription

    settings › cloud transcription › your server, type http://my-pc:8080, save, then test.

  2. Brain AI

    settings › brain AI: the server address (e.g. http://my-pc:11434), its key if it needs one, and a model name. test checks it and fills in a model if none is set.

my-pc ~ ap7 setup page 1 of 1

# transcription: whisper.cpp, Thai first

tools/whisper-server.sh

# Thonburian, a Thai fine-tune. Best for Thai, and Thai mixed with English

MODEL=large-v3-turbo tools/whisper-server.sh

# OpenAI's multilingual turbo, language auto. Better for English-only tapes

building whisper.cpp (CUDA: nvcc found)…
model → ~/.local/share/ap7-whisper
listening on 100.x.y.z:8080 (tailscale address only)

# brain AI: any OpenAI-compatible server. Ollama shown here

OLLAMA_HOST=0.0.0.0 ollama serve
ollama pull gemma4:e4b

# good Thai, a few seconds per summary on an RTX 3070

# llama.cpp, LM Studio or a LiteLLM proxy (with its key) work too

AP-7 asks for answers without "thinking". Several times faster on thinking models.

Where your tapes live

48kHz16bit1monoWAVplain

Every tape is a plain WAV file you can take anywhere. They're kept in the app's own folder; export copies one to the shared Recordings/AP-7/ folder, where My Files and a computer over USB can reach it.

  • Android/data/app.ap7/files/Recordings/
    • ap7_20261004_010745_team meeting.wav
    • ap7_20261004_010745_team meeting.transcript.json
    • ap7_20261005_081530.wav
  • Recordings/AP-7/ ← export lands here

The transcript sits beside its tape as JSON: the text, the engine and model, the time, and the latest brain AI result of each kind. Delete the tape and it goes too.

T01team meetingtxtai03:12
T02ap7_20261005_081530txt…00:47
T03saturday rehearsal…01:12

txt it has a transcript, ai it has brain AI results, … still working.

ap7_20261004_010745_team meeting.wav

The title goes after the timestamp, so it travels with shares and exports and tapes stay in recording order.

Built for the Galaxy Z Fold 8

Closed, the cover screen shows the face alone, held in portrait. It has nearly the proportions of a real recorder. Open it and the face sits beside a tape list that's always there.

closed: the face open: face + tape list txt ai txt
credits, where they belong

Where the idea came from

AP-7 is truly inspired by the TP-7 field recorder from Teenage Engineering. We love it so much, but there's no budget for one yet, so we built a recorder that lives in the phone while we save up. The credit for the idea is theirs.

Their gear is already on the desk here: I own an OP-1 and a few pocket operators.

AP-7 is an independent app. It isn't made, endorsed or sponsored by Teenage Engineering. TP-7, OP-1 and pocket operator are their names.

my OP-1 pocket operators TP-7, saving up

Privacy, in short

  • Tapes are recorded and kept on your phone.
  • Nothing is uploaded until you tap transcribe or a brain AI action, and then only to the service or server you chose.
  • Your keys stay on the phone and out of backups.
  • No account, no ads, no tracking.

Read the full privacy policy

Get an early tape

AP-7 is still on the bench. If you'd like to try it before it's out, write and say which phone you have.

Also on this tape