Data Ingestion
WeChat exports, SMS, voice notes, photos, video subtitles, diary files, and family corrections enter a guided local wizard.
Free stack: local parsers + Whisper/SenseVoiceMom passed away two years ago.
But last night, when I couldn't decide
whether to take the job in another city —
I called her.
And she picked up.
This is not fiction. This conversation happened last night.
Her mother has been gone for two years.
"We broke up three months ago. But at 2am when I can't sleep, I still hear him say goodnight. I know it's not really him. But it helps me breathe."
— Mia, 24
"I'm studying 8,000 miles from home. I bombed an exam and wanted to quit. I opened the app and heard my dad: 'Still up? Go to bed, kid. Tomorrow will be better.' I cried. Then I studied."
— Yuki, 20
"Grandpa is 87. Every Sunday I record our calls. One day my children will thank me."
— Haruto, 28
Choose a persona. Type or speak. Feel what Presence feels like.
🎤 Click mic to speak · Type and press Enter
An AI that learns everything about someone — their voice, their humor, the way they comfort you, the way they love — and preserves it forever.
You train it. You teach it. The more you talk, the more it becomes you. And when the time comes — it speaks for you. With your warmth. Your words. Your love.
Not just for goodbye — for every moment you wish they were here.
Their voice, their wisdom, their way of loving — doesn't have to disappear with them.
They're alive but gone. Hear their voice until you're ready to let go.
8,000 miles away, but dad's voice at 3am still works.
Record yourself now. They'll call the you of today when they need you most.
Leave yourself behind. So your family always has you.
Your voice doesn't have to disappear with your body.
Voice recordings, stories, photos. 10 minutes to start. A lifetime to build.
Link WeChat, WhatsApp, Instagram. It learns from real conversations.
1–3 people. A phone number is enough. Choose with love.
If you go silent, it checks in. No response — it reaches your person, gently.
A call in your voice. A message in your words. You're always there.
Pitch, pace, the way you say things differently to everyone. Not a robot — you.
Jokes, sarcasm, how you tease. Because humor is how we love.
How you comfort, how you react. The temperature only you carry.
Hear their voice, or read their words. You choose every time.
Every conversation makes it more you. It never stops.
They just talk. No setup needed.
Voice is only the mouth. Presence also needs memory, personality, emotion, data ingestion, trust control, and a real waitlist path. This upgrade makes every missing white-paper module visible and testable.
WeChat exports, SMS, voice notes, photos, video subtitles, diary files, and family corrections enter a guided local wizard.
Free stack: local parsers + Whisper/SenseVoiceSpeech habits, favorite words, relationship facts, values, taboos, and decision patterns become a private personality profile.
Free stack: LanceDB / sqlite-vec / local RAGEmotion2vec reads warmth, grief, stress, breath, hesitation, anger-as-care, and comfort timing from real audio.
Downloaded: emotion2vecLocal context retrieval supplies facts and style. The brain can run local models first, with DeepSeek only as optional analysis mode.
Privacy rule: raw family data stays localVoxCPM2 clones voice; MOSS-TTS-Nano gives CPU demo; MOSS-TTSD turns scripts into long emotional dialogue.
Downloaded: VoxCPM / MOSS familyActivity checks, trust contacts, delivery rules, revocation, and “do not impersonate living human” markers become real controls.
First build: local policy simulatorEach module maps to one part of making a preserved person feel human: not only what they say, but how long they pause, when they tease, when they scold, and when their voice softens.
“先喝口水，慢慢说。你不用在我这儿装没事。”
These repositories are now downloaded on the Mac under ~/presence-app/open-source-modules or already present under ~/presence-app/emotion. The site shows the actual role of each module instead of vague “AI magic”.
OpenBMB VoxCPM: 30-language voice design and controllable cloning. Presence uses it for the strongest free cloning route.
0.1B multilingual TTS for cheap local proof. This is the fastest path to “upload sample → hear result”.
Long-form spoken dialogue with multi-speaker control. This maps to family calls, interviews, and story reconstruction.
High fidelity, long-form speech, sound and streaming path. Used as the future voice backend family.
Extracts emotional representation from audio: grief, calm, joy, anger, stress, comfort.
Turns voice/video into text and supplies emotion cues for memory ingestion.
Local emotional dialogue experiments for comforting, non-generic responses.
Existing expressive voice module kept as fallback and comparison baseline.
The site now shows the actual architecture: local files become searchable memories, retrieved moments become context, and the answer is forced through the person’s style guard.
WeChat chat export, voice notes, photos, video subtitles, diary, family notes.
Remove duplicates, split speakers, label relationship, detect sensitive/private material.
Chunk memories into LanceDB/sqlite-vec. Nothing leaves the user’s device by default.
Find facts, tone examples, similar past conversations, emotional state, taboos.
Generate short, natural replies with style rules: “care first, advice second, never generic”.
Stream text into Nano/VoxCPM/MOSS-TTSD and play as the preserved voice.
Select the source material and privacy mode to see the pipeline Presence would run.
Affordable for the whole world.
Start preserving yourself.
Voice + all platforms + Guardian.
5 profiles for everyone you love.
End-to-end encrypted. Privately stored. Never used to train other models. Ever.
Be first to preserve someone you love.