How to import recordings you already have

The app and the web UI have no import button yet. Importing goes through the HTTP endpoint POST /api/chunks: one file per request, the body a whole audio file, decoded by ffmpeg on the server.

What you can import

What you have	Can it be imported	How
A voice recorder or phone recordings (MP3, WAV)	Yes	Upload as is
`.m4a` (e.g. iPhone Voice Memos)	Yes, after converting	A long `.m4a` can't be decoded from a pipe; run `ffmpeg -i in.m4a out.mp3` first
Plaud	Yes	Export MP3 or WAV in the Plaud app, then upload it
Anker × Feishu 录音豆	Tedious	The audio downloads only one recording at a time from the Feishu Minutes web page; download it, then upload it
OPPO 心力球	No	It keeps no original audio, so there is no audio to import

The sources for the third-party products are in What alldayrec is.

Rules

Use a device name of its own, e.g. import. A new name for each batch (e.g. import-2026-10) is the least hassle: the next rule holds per device name, so a new batch is transcribed even when it has recordings older than the last batch's.
Upload oldest first. Under one device name, audio earlier than where the server has already processed is kept but not transcribed.
After the last file, tell the server the batch is over. Otherwise the last stretch keeps waiting for more audio and never gets a final transcript.

Steps

Put the token in an environment variable:
bash
```
TOKEN=<your token>
```

Upload a file. start is the moment the recording began (Unix ms); it sets the day and time its conversations land on.

Linux (GNU date):

bash

curl -H "Authorization: Bearer $TOKEN" -H "Content-Type: audio/mpeg" --data-binary @rec.mp3 \
  "https://<your-server>/api/chunks?device=import&start=$(date -d '2026-10-20 09:30:00' +%s)000"

macOS (BSD date):

bash

curl -H "Authorization: Bearer $TOKEN" -H "Content-Type: audio/mpeg" --data-binary @rec.mp3 \
  "https://<your-server>/api/chunks?device=import&start=$(date -j -f '%Y-%m-%d %H:%M:%S' '2026-10-20 09:30:00' +%s)000"

On success, it returns the file's length in seconds, e.g. {"seconds":14.19}. A 400 means the server can't decode the file.

Upload the batch's other files in time order.
End the batch. t is the moment the last file ends (Unix ms), that is, its start plus its duration:
bash
```
curl -X POST -H "Authorization: Bearer $TOKEN" "https://<your-server>/api/chunks/end?device=import&t=<end-time>"
```
It returns {"ok":true}. Once the server has processed the remaining audio, the transcripts show up on that day's page, and the batch's last conversation ends: it no longer shows as 「进行中」 (in progress). A conversation with less than 10 seconds of speech is not kept as a conversation; its sentences stay in the transcript (see Architecture and data flow).

Content-Type

The Ogg Opus that the app sends uses audio/ogg; the server also accepts audio/webm and audio/aac. It stores other types (e.g. audio/mpeg for MP3, audio/wav for WAV) as they are, and ffmpeg detects the format itself.

Reference client

firmware/devkit/lr_sync.py is a Python client that reads a recorder's card and uploads it. The endpoint's parameters are in HTTP API.

How to import recordings you already have ​

What you can import ​

Rules ​

Steps ​

Content-Type ​