Heard Again is open source. You can download it, run it on your own machine, and never pay us a cent. That is not a loophole in our business model — it is the point. We wrote about why in why open source matters for your family's digital legacy, and nothing in this post walks that back.
But “you can self-host it” is not the same as “you should self-host it.” Most comparison posts written by a company that sells hosting are quietly rigged. This one tries not to be. Below is what each option actually asks of you, with real numbers, so you can pick the one that fits your household instead of the one that fits our revenue.
The Short Version
Self-host ifyou already run a home server, you are comfortable with Docker and backups, and the idea of your family's recordings living on someone else's hardware bothers you more than the maintenance will.
Let us host it if the person in your family who cares most about these recordings is not the person who would be maintaining the server — or if that is the same person and they already have enough to do.
That is genuinely the whole decision. The rest of this post is the detail behind it.
What Self-Hosting Actually Needs
There are two very different tiers here, and conflating them is how people end up disappointed.
The archive on its own— photos, documents, audio files, the family tree, sharing and consent controls — is modest. Four CPU cores, 8 GB of RAM, 20 GB of free disk to start, and Docker. That runs comfortably on a mid-range mini PC, an old desktop, or a decent NAS. If this is all you want, self-hosting is a reasonable weekend project.
Voice cloning and narration are a different animal.Generating speech in a relative's voice runs a text-to-speech model locally, and that wants eight or more CPU cores, 32 GB of RAM, an NVIDIA GPU with 24 GB of VRAM — an RTX 3090 or 4090 class card — and another 50 GB of disk for models and generated audio. That is not a casual requirement. A used 3090 is still several hundred dollars, and it will draw real power whenever it works.
There is a middle path: run the archive locally and point the voice features at a cloud endpoint, so you keep your files on your own disk and rent the GPU only when you generate something. Heard Again supports that split directly — you choose local, cloud, or hybrid for compute and for data separately when you register a self-hosted instance.
Worth saying before you agonise over it: whichever side you land on, you are not locking the data in. The tree exports to GEDCOM, which every genealogy program reads, and the audio exports as ordinary files. The one thing that does not travel cleanly is the link between the two — GEDCOM carries people, not recordings — so if you expect to move between setups, read what survives a GEDCOM import and what you have to reattach before you build a filing habit you will have to undo.
“Self-hosting is free in money and expensive in attention. Hosting is the reverse.”
The Cost Nobody Prices In
The hardware is the easy part to budget. The hard part is the standing commitment. A self-hosted archive needs someone to apply updates, watch the disk fill up, renew certificates, and — above all — verify that the backups restore. Not that backups run. That they restore. An untested backup is a rumour.
Then there is the question most home-server guides skip: what happens to this machine when you are not able to look after it any more? A family archive is meant to outlive the person who built it. If the recordings live on a box in your basement, and only you know the disk layout and the admin password, you have built a single point of failure with your family's memories inside it. The practical rules for preserving family memories digitally — more than one copy, in more than one place, labelled so a stranger could understand them — apply to self-hosters most of all, because nobody else is applying them for you.
None of this is an argument against self-hosting. Plenty of people run exactly this kind of setup well. It is an argument for being honest with yourself about whether you are one of them, before the archive matters.
What You Give Up by Paying Us
Fairness cuts both ways, so here is the honest list. On a hosted plan, your files sit on our infrastructure rather than yours. You are trusting our access controls and our operational discipline. You are accepting a storage allowance instead of “however big a disk I feel like buying.” And you are taking on a recurring bill, which is a different kind of commitment than a one-off hardware purchase.
What stays the same either way: the software is the same software, the consent controls are the same controls, and your data is exportable. Because the project is open source, a hosted plan is not a trapdoor. If you outgrow us or stop trusting us, you can take the export and run the same stack yourself. That is the part that should make paying us feel safe rather than binding.
When Paying Is Simply the Right Answer
Here is the plain reason to pay, with no marketing on it. The recordings you most want to keep are usually the ones you have the least time to look after — because the reason they are urgent is that someone is ageing, or ill, or you have just found a box of tapes and the family is already asking questions. Standing up a server is the wrong task to take on in that week. The plan you can actually start today is $9.99 a month, with voice generation included — thirty minutes of generation a month and up to fifty voice profiles — and a 14-day free trial before anything is charged. There is a cheaper $4.99 storage-and-sharing tier with no AI features listed on the pricing page, but it is not open for signup yet, and we would rather tell you that here than let you find out at the checkout. Either figure is less than the card you would need to buy, and comparable to what an always-on machine with that card in it adds to a power bill. You can compare the plans on the pricing page. You are buying the backups, the updates, and someone to email when it breaks.
And you can change your mind later. Start hosted while the recording matters most, export and self-host once things calm down. Or run your own instance and keep a hosted copy as the off-site backup you were never going to get around to building. Those are both sensible.
Decide the Hosting Later, Record Today
One last thing, and it is the most important paragraph here. This whole decision is reversible. The conversation you have not recorded yet is not. If you are weighing hosting options while there is a relative whose voice you have never captured, you are optimising the wrong step. Go record the conversation on your phone first. Sort out where it lives afterwards. The file can be moved. The afternoon cannot be rescheduled.