Bring your own model
A dozen providers built in, one-tap switching and balance checks. If the main model goes down or runs out of credit, it fails over to a backup.
- DeepSeek
- Doubao
- Qwen
- Kimi
- Zhipu
- OpenAI-compatible
- Self-hosted
Listening
Your 24-hour life, as context.
All you need is your Android phone. It listens all day, transcribes, recognizes who is speaking, and lets AI turn it into todos, promises and memories. No hardware, no account, and your data never leaves your phone.
v0.5.5 · 105 MB · Android 10+ · Served from the ModelScope mirror
Send Wang the quote before next Wednesday — oh, and the keys are in the second hallway drawer.
TaoSheng filed
Transcribed on device · speaker recognized
24 hours
You never have to remember to open it. It records only when someone speaks, and stays quiet otherwise. Scroll through one day with TaoSheng.
08:40
On-device voiceprints recognize you, your family and your coworkers. Mom’s line at breakfast is labeled “Mom”. Not recognized yet? Long-press the line, pick “who is this” once, and it learns.
11:05
Tap “Topic” for a meeting, class, call or script. It organizes as you go, ten minutes at a time, so notes are ready soon after you wrap. Tap “Mark” at the important bits for extra detail.
14:32
“Send Wang the quote before Wednesday” becomes a todo with a reminder. Later you say “sent it” and it ticks itself off. Important ones can ring like an alarm, full screen on the lock screen.
19:20
“Who has my power bank?” “What did I promise Wang last week?” TaoSheng searches your memory with real query tools, and every answer cites the exact line. Tap to hear it again.
23:41
Leave the phone by your bed and it starts when you fall asleep. A sound map marks snoring, coughs and sleep talk — the talk is transcribed, and every dot replays. Unlock, unplug or dismiss the alarm and it stops.
07:05
Where you went, who you talked to, which apps you used and how you slept — finalized before dawn as “the day’s trail” and delivered to your notes, cloud drive or AI agent.
The day’s trail · Nov 20
→ Delivered to WebDAV · Notion · Obsidian · your agent
Privacy
TaoSheng has no accounts and no server of ours. By default, nothing leaves your phone.
0
accounts
Install it and start. No sign-up.
0
servers of ours
We never receive or see any of your data.
3
days until raw audio is deleted
Adjustable from 1 to 365. Starred clips are kept.
100%
on-device voice detection & voiceprints
Transcription can run fully offline.
On your phone
“Delete the last N minutes” and “don’t record this” erase it for good.
Only if you turn them on
Data only goes to services you choose, with your own keys. Pick local transcription plus a self-hosted model on your LAN and nothing leaves your own devices.
Hardware
Take apart an AI recording pendant and you’ll find a Bluetooth mic and a battery. TaoSheng can use any Bluetooth audio device as its source — hang your earbuds on a string and you have one. When they die or disconnect, it falls back to the phone mic.
It can also pick up your phone’s own call recordings, with who and when. It never reads your contacts and never sends numbers to any model.
More
A dozen providers built in, one-tap switching and balance checks. If the main model goes down or runs out of credit, it fails over to a backup.
Each day is written as Markdown + JSON. Tap “Copy for agent” in the sync folder and paste into Claude Code or Codex — it gets your 24-hour life as context, read-only.
$ claude› What did I promise yesterday?Reading TaoSheng/2026-11-20.md …You promised Wang the quote by Wednesday; Lin will return your power bank tomorrow.
Reviews, todos, skill cards and daily memory docs, pushed incrementally. No changes, no network.
WebDAV, Nutstore or any S3-compatible storage, passphrase-encrypted with 3/7/30-day versions. New phone? Restore everything, settings and keys included.
Meeting notes, end-of-day check-in, weekly review and mood journal out of the box. A JSON file plus a prompt is a new skill.
Silero VAD records only speech: about 0.5% per hour in testing. Interrupted transcription resumes on its own, with keep-alive guides for Xiaomi, Huawei and Honor.
No. The app is free and the source is public. Cloud transcription and LLM calls are billed by your own providers; with on-device transcription, that part costs nothing.
No, the phone mic is enough. Lavalier mics, Bluetooth earbuds and DJI Mic switch automatically when connected.
Android 10+ on 64-bit (arm64) phones, which covers almost every recent phone. It is mainly tested on a Xiaomi 14 Pro, with keep-alive guides for Huawei, Honor and HarmonyOS.
No — we don’t have servers. Data only leaves your phone when you turn on cloud transcription, an LLM, push channels or backup that you configured yourself.
Transcription can, with the on-device SenseVoice model. The digest needs an LLM, but you can point it at a self-hosted model on your LAN.
It records only when someone speaks: about 0.5% per hour in testing. Sleep tracking reminds you once if you’re below 30% and not charging.
From 0.4.0 it is FSL-1.1-MIT: personal, educational, research and internal company use are all fine; building a competing commercial product is not. Every release becomes MIT two years after it ships. Versions up to 0.3.1 are MIT.
Free to download, no sign-up. All you need is an Android 10+ phone.
v0.5.5 · 105 MB · Android 10+ · Served from the ModelScope mirror
Open in your browser
This in-app browser cannot download APKs. Tap ··· in the top-right corner and choose “Open in browser”.