I’m building an AI document archive after work and after the kids go to sleep. It’s been live 2 weeks and I still have 0 users.
Hey IH,
I’ve been reading Indie Hackers for a while and finally decided to post.
I’m from the Czech Republic and I work full-time as an IT infrastructure engineer. I’ve got a wife, two kids, and I’m building Veluvanto in the hours that are left after work and family. Some evenings I get two decent hours. Some evenings I open the laptop, fix one small thing, and that’s it.
Veluvanto came from a very ordinary problem.
One evening I needed a warranty receipt for something we wanted to return. I knew I had it somewhere. Maybe in email, maybe in cloud storage, maybe in some folder on my PC from two years ago. I found it after maybe 20 minutes, but that was exactly the annoying part. Nothing dramatic, just one more small task eating the little free time you have left at the end of the day.
That got me thinking about document management tools.
On one side you’ve got enterprise systems like DocuWare or M-Files. Powerful, sure, but also expensive, heavy, and made for companies that are willing to spend time on setup and process design. On the other side there’s Paperless-ngx, which I actually respect a lot, but it’s still self-hosted and not something most normal families, freelancers, or small business owners will want to run.
I felt like there was a gap there:
people who don’t want a giant enterprise DMS,
but also don’t want to self-host and babysit another tool.
That’s why I started building Veluvanto.
The basic idea is simple: upload documents and let the system do the sorting and extraction for you.
It supports the boring real-world stuff people actually have:
PDFs, scans, phone photos of receipts, Word docs, spreadsheets, presentations, TIFFs.
For files that already contain text, the text is extracted locally first. For scans and image-based files, AI vision is used. Internally the content is converted into a format the model can work with properly, so the document structure isn’t lost. Then AI pulls out useful fields like dates, amounts, invoice numbers, parties, tags, document type, and so on.
The point is that you shouldn’t need to build a folder tree or invent a tagging system before the app becomes useful. You upload a file, and it should already be organized enough to find later. If the AI gets something wrong, you can still override it.
I’m also building a few features around that core:
- AI assistant for asking things like “show me unpaid invoices from Q1” or “summarize this contract”
- Search with citations, so the AI has to point back to the source document
- Approval flows that are simple enough for small businesses, not just enterprises
- Email ingestion, so documents can land in the workspace automatically
- EU-hosted infrastructure and isolated tenant storage, because I care a lot about data separation and operational sanity
That last part is probably the infra engineer in me.
I actually enjoy the operational side of SaaS. Not just building features, but also thinking about isolation, backups, monitoring, recovery, routing, and keeping the thing stable long term. A lot of people hate that part. I kind of like it.
The stack is pretty boring on purpose:
Go on the backend, PocketBase embedded, SQLite per tenant, React 19 on the frontend, OpenSearch for search, Docker Compose over SSH for deployment, Hetzner for compute, Backblaze B2 for storage, Gemini through LiteLLM for AI.
No Kubernetes. No “future-proof” complexity. Just stuff I can actually run myself.
Current pricing is:
free tier with no credit card,
then €9 / €29 / €99 per month depending mostly on storage and AI usage.
Where I am right now:
the product has been live for about 2 weeks
and I still have 0 users.
I’ve submitted it to a bunch of directories, I’m working on SEO, and I’ve spent more time fixing launch bugs than I wanted. I don’t have an audience, no newsletter, no Twitter following, no built-in distribution. I’m just trying to get the first real people to use it and tell me where it’s bad.
The product is here:
https://veluvanto.com
So my question is mostly for other people who built boring SaaS products without an audience:
How did you actually get your first 10 users?
Not the polished advice version. The real version.
What did you do, where did they come from, and what ended up working?
Veluvanto is doing too much too early.
The core problem is simple and painful:
“find the document fast.”
That’s the wedge.
But the product is already talking like a full document operating system:
AI assistant, approval flows, email ingestion, isolated tenants, search citations, infra design.
That’s a lot of surface area before the first user has even successfully solved one annoying 20-minute retrieval problem.
Right now the risk is not distribution.
It’s that the product is broader than the first pain.
People don’t buy “AI document archive.”
They buy:
“I need that invoice in 10 seconds.”
That’s the first job.
Everything else is second-order.
The first 10 users usually come from selling the narrowest painful use case manually:
freelancers who lose invoices
small agencies chasing receipts
people handling reimbursements
parents buried in warranties / tax docs
Not “try my AI archive.”
More like:
“Upload 20 messy receipts and see if this finds the one you need in 10 seconds.”
That gets tested faster.
And honestly, Veluvanto is the bigger issue here.
It sounds elegant, but too abstract for a product people need to trust with contracts, invoices, and personal records.
For this category, the name has to feel simpler, cleaner, and more immediately dependable.
Lyriso.com would carry this much better if the product stays document-first and trust-heavy.
Building Veluvanto while juggling a full-time job and family is a massive feat, and that document search frustration is a pain point we all share. Since you're starting from zero, finding people in "self-hosted" forums who are tired of managing their own servers could be your fastest win. It’s usually about joining the right conversation at the right time instead of waiting for SEO. I really hope you get some good signals soon for both of us, of course!
Thanks for tip. But finding those and targeting is not easy job. Because when they are tired of selfhosting, they do not report to selfhosting topics due to possible hate from other selfhosters, who do not like saas.
That makes total sense! I completely agree with your point.
As someone who is also juggling work and family life, I recently launched a product called bunzee.ai. It’s built to drastically cut down the painful research time for makers who are stuck in 'Excel hell' with dozens of tabs open just to do market research and planning.
We’ve officially launched, but to be honest, we are still struggling through the cold-start phase. If you’re open to it, I would love to get your brutally honest feedback on bunzee.ai! I’d also love to find a way to return the favor and help you out as well.
I took a quick look and went through the flow (signup → onboarding → dashboard).
First impression — the idea makes a lot of sense, but the first “value moment” feels a bit delayed.
Right now the experience is:
sign up → answer a few setup questions → land on an empty dashboard
At that point I wasn’t fully sure what to do next or how to quickly experience the benefit.
I ended up looking for something like:
– a sample document already inside
– or a super quick way to test how extraction + search actually works
Because the core promise is very strong (“everything is instantly findable”),
but I don’t really feel it until I upload and interact with real data.
Also small UX thing:
there are a couple of slightly confusing moments early on (e.g. duplicate actions leading to the same place, tutorial state not being very clear), which adds a bit of hesitation.
Overall it feels like the value is there,
but it takes a bit too much effort to reach that “aha, this actually saves me time” moment.
Curious how you’re thinking about making that first interaction faster and more obvious.
Thank you, your feedback means a lot to me. My idea behind tutorial is to provide user real experience, not just some fake demo with fake extraction, but real. I will try to make more clear to reach wow effect faster as you suggest. I fully agree, empty dashboard at the start is not really good. I know, I do have developer blindness for some things :) So any feedback is more than welcome
That makes sense — and I think your instinct about “real experience vs fake demo” is right.
The tricky part is that new users don’t really think in terms of “real vs demo” at that point.
They’re just trying to answer:
“do I understand this and is it useful for me?”
So even a real flow can feel unclear if the first step requires too much effort before showing value.
What might help is not replacing the real experience,
but compressing it into something like:
– one guided action
– immediate visible result
– then let the user continue naturally
Right now it feels like the value is there,
just slightly too far from the first interaction.
If helpful, I can map out a quick “first 2-minute experience” for a new user
and where exactly the hesitation points are — I already saw a couple while going through it.
I investigated it and my steps will be:
That sounds like a solid direction — especially making the tutorial faster.
The ghost overlay is interesting, but one small thing I’d be careful about:
it shows what the dashboard will look like,
but it doesn’t let the user experience why it’s useful.
So it might reduce the “empty feeling”,
but not necessarily create that “oh, this actually saves me time” moment.
Feels like the strongest version of this would be something the user can interact with immediately, not just see.
For example:
even a single pre-filled document or a quick guided upload → result → search loop
might make the value click faster than a visual preview.
Right now it seems like you’re very close — it’s mostly about turning that first interaction into something tangible, not just visible.
Happy to sketch what that could look like in practice if helpful.