organ · 01
● head → mind
Mind.
How you think — reflection, openness, formality — traced across cognitive bands.
reflection0.91
openness0.76
formality0.30
a personal ai
·
lives on your phone
Meet your
Twin
grown from your voice.
DailyVox listens, transcribes, and quietly builds a model of your personality — every word of it running on-device. No cloud, no accounts, no one watching. Just you, observed by yourself.
0rpc
network
9
frameworks
∞
private
$0
forever
subject · self
id · twin_v1.1
captured · on-device
entries · 142
voice_entry.live · recording on-device
SFSpeechRecognizer → NLTagger → TwinEngine
sentiment
+0.00
entities
—
Twin
idle
network
0 B sent · local only
meet your Twin
An anatomy of your Twin.
Your Twin isn't one model — it's four, each one reading a different region. Mind for how you think, voice for how you speak, heart for how you feel, and a graph of who and what you mention. Below it all, a forecast of where you're headed.
organ · 02
● throat → voice
Voice signature.
How you speak — rhythm, tone, pace — measured as a live acoustic fingerprint.
● warm
152 wpm
pauses · low
expressiveness0.68
directness0.55
live vitals
mind
0.91
voice
0.68
heart
+0.42
graph
42
14:22:28 ✓ boot · on-device · 0 rpc
14:22:30 ▸ ingest entry #143 · 02:14 audio
14:22:31 ▸ twin.update · valence ▲ +0.04
14:22:33 ✓ persist AIState · 4.2 KB · local
organ · 03
heart → emotion ●
Emotional signature.
How you feel — valence, arousal, dominance — as a breathing field of colour.
valence+0.42
arousal+0.28
organ · 04
body → graph ●
Knowledge graph.
Who and what you mention — weighted by how you feel when you mention them.
entities42 nodes
top-mentionSarah ×18
organ · 05 · forecast
Tomorrow's mood.
tomorrow · +0.58 · leaning positive
TUE
+.58
WED
+.44
THU
+.38
FRI
+.62
SAT
+.74
SUN
+.68
MON
+.50
how it learns
From voice to Twin, five stages.
Every journal entry flows through the same on-device pipeline. Click any stage to see what happens to your data at each step. (Spoiler: nothing leaves the phone.)
stage 01 · capture
Microphone to raw audio, locally.
AVAudioEngine captures AAC audio at 44.1 kHz directly from the iPhone
microphone. The buffer is held in memory and handed to the transcriber —
never uploaded, never cached to disk.
- framework
- AVFoundation
- format
- AAC · 44.1 kHz
- latency
- < 10 ms
- network
- 0 rpc
live view
what leaves your phone
Nothing. On purpose.
Attach a network proxy during a 2-minute voice entry. Run the transcription, NLP, and Twin update. Zero bytes leave the device. This isn't a promise — it's the architecture.
0rpc
network calls
Every AI operation runs on Apple's Neural Engine. Zero outbound requests.
0sdk
third-party
No analytics, crash reporting, or ad SDKs. Apple frameworks only.
AES-256
encryption
GCM authenticated encryption for backups. Keys in the Secure Enclave.
—
account required
Data Not Collected — Apple's strictest privacy label.
the stack
Nine Apple frameworks. Zero vendors.
DailyVox is built entirely on Apple's on-device SDKs — no third-party dependencies anywhere in the app. Every capability has a home inside iOS.
Sp
SFSpeechRecognizer
Speech.framework
On-device transcription with requiresOnDeviceRecognition = true. 60+ languages on the Neural Engine.
Nl
NLTagger
NaturalLanguage
Sentiment scoring, Named Entity Recognition, part-of-speech, and language identification — four CoreML models running locally.
Ne
NLEmbedding
512-dim vectors
Dense sentence embeddings for semantic search and k-means thematic clustering of your journal.
Fm
Foundation Models
3B on-device LLM
LanguageModelSession with tool calling, @Generable structured outputs, streaming.
Sa
SpeechAnalyzer
next-gen ASR
Replaces SFSpeechRecognizer in v2.0. Lower latency, native long-form support, no permission prompts.
Cd
Core Data + CloudKit
NSPersistentCloudKitContainer
Local SQLite with optional end-to-end-encrypted sync. The Twin is one AIState Codable JSON entity.
Ck
CryptoKit
AES-256-GCM
Authenticated encryption for backup exports. Passphrase-derived keys for device migration.
La
LocalAuthentication
Face ID · Touch ID
Biometric gating on every journal access. Keys stored in the Secure Enclave — never extracted.
Wk
WidgetKit + AppIntents
widgets · siri
Home & Lock-screen widgets. AppIntents-powered Siri shortcuts for hands-free entries.
next to the category
How it compares.
| DailyVox | Day One | Reflectly | Apple Journal | |
|---|---|---|---|---|
| price | free forever | $34.99/yr | $59.99/yr | free |
| voice journaling | ✓ | — | — | — |
| on-device AI | ✓ | — | — | partial |
| Twin · personality model | ✓ | — | — | — |
| works offline | ✓ | partial | — | ✓ |
| data collected | none | yes | yes | minimal |
| account required | no | yes | yes | no |
| open source | ✓ | — | — | — |
note from the builder
“I wrote a diary for twenty years. Then I couldn't anymore. So I stopped writing — and started talking. That's how DailyVox was born, and why it runs entirely on your phone instead of someone else's server.”
Every voice-journaling app I could find shipped audio to a cloud for transcription. I didn't want my most private thoughts processed by a third party, and I didn't want to pay a subscription for it either. So I built my own — free, open-source, fully on-device. The Digital Twin is the next step: a model of you that lives on your phone, not on somebody's server. Read the full story →
from the journal
Writing about the work.
2026.04.01 · technical
How your Twin learns your personality.
Inside the TwinEngine — how NLTagger signals become a multi-dimensional personality model, on-device.
read →
2026.04.01 · privacy
Why your AI journal should work offline.
If the app needs the cloud to transcribe, it isn't private. The on-device test — and how to run it yourself.
read →
2026.04.01 · protocol
The ADHD voice-journaling protocol.
Speaking at 150 WPM versus typing at 40. A protocol for brains that think faster than their fingers.
read →
install · $0 · no account
Start talking.
Your Twin is listening.
Two minutes a day. No writing. No subscription. No data collection. Your voice, your personality model, your phone.
Download free on the App Store →