Public Record
Provenance
EchoVault shipped text, voice, and real-time video conversation as a single digital legacy product in June 2025, built by one person. This page is the record of when that happened and where each date is documented. Every entry below links to a source you can open and check for yourself, and to an Internet Archive capture in case the source changes.
Sources verified and archived 23 August 2026
The record
The EchoVault project page went live on Devpost as an entry in the World's Largest Hackathon presented by Bolt. The byline on that page reads 29 June 2025, 11:28 PM EDT. The Internet Archive captured the same page on 26 July 2025, so the writeup described below exists in a third-party copy made weeks after it was written.
The web app went live at echovault.me with all three conversation modes running: text, voice, and real-time video. The Devpost submission lists echovault.me in its “Try it out” section as the working demo, which is how the entry was judged. At that time the root domain served the application itself; the marketing site now there came later, and the app moved to app.echovault.me.
EchoVault won the Inspirational Story bonus prize at the World's Largest Hackathon presented by Bolt, an event with more than 130,000 entrants. The award is recorded on the Devpost project page and the project carries a winner ribbon in the hackathon's public gallery. Both are on Devpost rather than on anything we control.
The iOS app was released on the App Store under developer of record Ugochukwu Nwune. It is an iPhone app requiring iOS 15.1 or later, and the web app continues to operate alongside it.
What the June 2025 writeup says
These are quoted from the Devpost submission as written in June 2025. Each one is present in the Internet Archive capture from 26 July 2025, which predates every product described further down this page.
- Three modalities, split by tier
- “I added voice and video avatars for paid subscribers and left the echo text system free to use.”
- Recurring check-ins, not one sitting
- “The user is required to have a check in at least once every 2 weeks, this is another core aspect of the project.”
- Release to appointed custodians
- “If a user does not check in 6 times in a row, it automatically triggers a deceased protocol that releases the user's echo to whoever they appointed during their onboarding process.”
- Inherited time already paid for
- “The custodians can interact with the echo for free for the duration that user was subscribed for, after which they will have the option to extend the subscription by paying themselves.”
The writeup also names the components used: Gemini for the language model and embeddings, ElevenLabs for voice, and Tavus for the video avatar. None of these were unavailable to anyone else in June 2025. Read the 2025 capture or the 2026 one.
The archived build
On 22 August 2025 the Internet Archive crawled echovault.me and saved more than the page. It saved the compiled JavaScript the application was running that day. What follows is quoted out of that file, which the Archive holds rather than we do.
- A deceased_protocol table, initialised with missed_checkins, last_checkin_date and contacts_notified fields
- Interface copy reading "Regular Check-ins" and "Every 14 days"
- Interface copy reading "If you miss 6 consecutive check-ins, your designated contacts will be notified"
- The tagline "Your professional digital biographer for meaningful life check-ins"
- Calls to elevenlabs-train-voice and elevenlabs-synthesize, writing to a voice_models table
- Calls to tavus-generate-video, writing to a video_avatars table
The Devpost page describes the system. This is the system as compiled and served, eight weeks after launch and three months before the earliest competitor described below began development. The archived page and the archived bundle are both public.
What existed before
Digital legacy products before 2025 were archives with playback. StoryFile filmed people answering hundreds of questions on camera and retrieved the closest matching recorded clip when a family member asked something. HereAfter AI ran a chapter-by-chapter audio interview and played the actual recording back, so what a family heard was the person's real voice. Both approaches shared a limit: they could only answer a question the person had already been asked on the day of recording.
That limit was described publicly in 2024, before we shipped. Systems of this kind are not generative the way a living person is, so heirs cannot come to one and get the benefit of the person's expertise by asking a question it was never asked. NorthBay biz, August 2024
StoryFile filed for Chapter 11 in May 2024 in the Southern District of New York, declaring roughly $1.5M in assets against $10.5M in liabilities, and emerged in February 2025 under new ownership. Westfair coverage and the docket are both public. HereAfter AI replaced its homepage with a shutdown notice during 2026. StoryFile's current site still positions its approach as capturing real people answering real questions rather than generating synthetic personas. storyfile.com, archived August 2026
Voice cloning, video avatar rendering, and retrieval-augmented generation were all off-the-shelf well before June 2025. The established products had the same access to them that a solo builder working through a hackathon weekend did. What had not been assembled was the product.
What happened next
That shape is now the common one. Products that launched after ours sell the same three things on their front pages: guided check-in capture rather than a single sitting, echoes that answer in text, voice, and video, and credits that pass to the family. A product matching the architecture closely, built on the same off-the-shelf components, began development in November 2025 and launched publicly in June 2026. A separate product organised around inherited prepaid time launched in January 2026. A company founded in 2024 rebranded late in 2025 and moved out of posthumous legacy entirely, having found that most of its users were not preparing for death.
We are not naming them and we are not suggesting anyone took anything. Each of those dates is checkable against the companies' own launch material, and readers who want to check them should. The only date this page asserts is ours, and it is the earliest one.
A note from the founder
I have no professional standing in grief or estate work. What I have is a record of calling this product's shape early. In June 2025 I decided the corpus had to be built through short recurring check-ins rather than one long interview, that families should inherit time already paid for, and that one product should carry text, voice, and video together rather than pick one of them. The category has since converged on all three. The full stack for this existed for years before I put it together, and the incumbents had it and did not.
All three of those decisions are quoted from the June 2025 writeup further up this page.
Where it stands now
EchoVault is still built and run by one person. The web app and the iOS app run on the same system, and the check-in model described on the Devpost page in June 2025 is the one still in use.
The open problem is what happens to an Echo whose creator never authorised anyone to export it. That work is in progress, and this page will be updated with dates as it lands.