Questions &
Answers.

Still With You is an AI system that preserves a loved one's voice, memories, and personality so family members can continue having natural conversations with them. It runs entirely on a local device in your home -- no cloud, no subscriptions, no data collection. Think of it as a bridge between love and technology: their voice, their stories, their humor, available whenever your family needs them.

For any family that wants to preserve the presence of someone they love. Parents who want to leave something meaningful for their children. Families who have already lost someone and wish they could hear their voice again. Grandparents who want their grandchildren to know them. Anyone who understands that some voices are too important to lose.

Both. If you are planning ahead, you can record voice samples, enter memories, and prepare milestone messages for the future. If you have already lost someone, you can still build their presence from existing recordings (phone calls, voicemails, home videos) and family memories. The voice cloning works with any clear audio of their voice, and memories can be entered by anyone who knew them.

Fundamentally different in four ways. First, it runs 100% locally on your device -- no data ever touches a server. Second, it speaks with their actual cloned voice, not a generic AI voice. Third, it has a persistent memory system that remembers family stories, relationships, preferences, and personality traits. Fourth, it recognizes family members by face and adjusts how it communicates based on who it is talking to and how old they are. It is not a chatbot. It is a preserved presence.

Still With You uses XTTS v2, a zero-shot voice cloning model. You provide 30 minutes of natural speech recordings -- not scripted readings, just them talking normally. The system extracts reference clips and learns their unique cadence, inflection, and emotional tone. When the system speaks, it synthesizes new speech in their voice in real time, preserving the warmth and personality of how they actually sounded.

About 30 minutes of natural speech. This does not need to be a single continuous recording. Multiple shorter recordings work well -- conversations, stories, even voicemails. The key is variety: different emotions, different topics, different energy levels. The more natural and varied the recordings, the better the voice clone captures their full range.

The voice will be recognizably theirs -- the cadence, the warmth, the way they pause before something important. Family members consistently recognize the voice as belonging to their loved one. It is not a perfect copy of every nuance, but it captures the essence of how they sound. For most families, the first time they hear it, they know immediately who it is.

Yes. You can add more recordings at any time to improve the voice quality. The system re-processes all available audio to create an updated voice model. More recordings generally means better quality, especially if the new recordings capture speech patterns or emotions that were missing from the original set.

If you have any audio at all -- phone voicemails, home videos, video calls, even social media videos -- the system can work with it. The audio does not need to be studio quality. If you truly have no recordings, the system can still function as a text-based presence with their memories, personality, and knowledge. Voice can be added later if recordings are found.

No. Zero data leaves your device. Every component -- voice synthesis, language model, memory database, facial recognition -- runs locally on the hardware in your home. There are no cloud servers, no API calls, no telemetry, no analytics. The device does not even need an internet connection to function. Your family's conversations are physically incapable of being transmitted anywhere.

No. We have no mechanism to access your data even if we wanted to. There are no remote access tools, no backdoors, no phone-home features. The software runs entirely on your device. We do not operate servers that store or process customer data. Your system is self-contained and we have no connection to it after delivery.

Nothing changes for you. Your device continues working exactly as it does today. All software runs locally. There are no license servers to check, no activation keys that expire, no cloud dependencies. The entire open source stack (Ollama, Weaviate, XTTS, etc.) is independently maintained by large communities. Your device is yours, permanently.

The data on the device is stored on an encrypted filesystem. This means if someone physically steals the device, they cannot read the data without the decryption key. During normal operation, the data is accessible only to the software running on the device. There is no network exposure -- no ports are open to the internet.

The device has no open ports to the internet and does not need a network connection to operate. This eliminates the most common attack vectors. For local network security, the web interface runs on your home network only. Facial recognition with IR liveness detection prevents spoofing with photos or screens. The attack surface is intentionally minimal because the device was designed to be offline-first.

All tiers are software-based -- you provide the hardware. The system requires 24GB of GPU VRAM for the full AI stack. The recommended setup is a mini PC with OCuLink support (e.g. Minisforum AtomMan X7 Ti), an OCuLink GPU dock, an NVIDIA RTX 3090 (24GB, used/refurbished works great), 32GB RAM, 512GB SSD, a USB webcam, and an external speaker. The Blueprint tier includes a verified hardware list with exact models and purchase links, a video build guide, and a setup call. Estimated hardware investment: $1,900 - $2,700.

GPU: NVIDIA RTX 3090 with 24GB VRAM (required for full AI stack). RAM: 32GB recommended. Storage: 512GB SSD. OS: Ubuntu 22.04 LTS or later. Camera: any USB webcam (IR-capable recommended for anti-spoofing). Speaker: external speaker or monitor with built-in audio. The GPU is the most important component -- it handles voice synthesis, language processing, and speech recognition simultaneously.

With the USB installer (Blueprint and Legacy tiers): plug the USB into your hardware, boot from it, and the setup wizard walks you through the rest. The personal setup (voice recording, memory entry, facial enrollment) takes 1-2 hours. A 45-minute video setup call is included with The Blueprint and Legacy tiers. For The Seed and The Voice tiers: install the software on your own Ubuntu system -- we provide video tutorials for every step. The Legacy tier includes lifetime phone and FaceTime support.

No. The device works completely offline. You may want internet for initial software updates after unboxing, but once set up, the device never needs to connect to the internet again. All AI models, voice synthesis, memory storage, and facial recognition run locally on the device's GPU.

Yes. The system was built by a father for his own children. It includes a parental alert system that monitors conversations for sensitive topics (self-harm, bullying, stranger interactions, substance mentions) and immediately notifies the surviving parent. For crisis situations like self-harm mentions, the AI is bypassed entirely and a pre-written, loving parental response is delivered instantly. All content is age-appropriate and adapts to each child's developmental stage.

Facial recognition. Each family member enrolls by looking at the webcam for a few seconds during setup. After that, the system instantly recognizes who walks up to the device and adjusts its personality, memories, and communication style accordingly. IR liveness detection prevents spoofing with photos or screens. No passwords or wake words needed -- just walk up and start talking.

A monitoring system that watches conversations with children for concerning content. If a child mentions bullying, self-harm, strangers, substances, or inappropriate topics, an alert is created for the surviving parent. Alerts are color-coded by priority (red for critical, orange for high, blue for medium) and displayed when the parent is identified by facial recognition. Critical alerts (self-harm) trigger an immediate pre-written crisis response before any AI involvement.

Yes. The system includes specialized Math and Coding tutoring modes. When a child asks a math question, the system automatically switches to a dedicated math model (Qwen 2 Math) that handles everything from basic arithmetic to calculus. For programming questions, it switches to a coding model (Qwen 2.5 Coder). Both modes include smart continuation detection so the tutoring session feels natural, not robotic. It explains concepts the way a patient parent would.

All tiers are estimated for October 2026. Because every tier is software-based (no hardware sourcing or assembly), there is a single delivery window. Physical items (voice kit cards, USB installer) have short manufacturing lead times. We will provide monthly updates throughout the fulfillment process.

Kickstarter pledges follow Kickstarter's standard refund policy. Before the campaign ends, you can cancel your pledge at any time. After the campaign ends, refunds are handled on a case-by-case basis. We are committed to delivering on every promise and will work with backers to resolve any concerns.

Yes. The Seed tier is fully digital. The Voice, Blueprint, and Legacy tiers include lightweight physical items (voice kit cards, USB installer) shipped via USPS. Shipping is free in the US; international shipping is $8-15 depending on region.