A platform this complete
should not be this small.
askFinz reads the web itself and runs 29 workspaces behind one sign-in, 13 of them shipping today. It was built by a small team on a small fleet, which is the point: the hard part is done, and what remains is reach. This page is for the people who can change that — companies, investors, institutions, accelerators and funding programmes.
These are not slide figures. They are the running system, read live, and they move while you look at them — including downwards on a bad day.
A web page kept whole — its markup, scripts, styling and navigation furniture — is vastly larger than the part anyone would ever search. askFinz keeps the cleaned text and the vectors and throws the rest away, which is why the figure above is measured in kilobytes rather than megabytes.
That is the whole reason the index can keep growing on hardware rather than on a balance sheet. The next million pages are a question of adding machines to a pool, not of signing a larger contract with whoever owns the data.
Filing a page is not the same as keeping it. Each one is scored, cut into passages, placed in a subject collection and described — so the index can be filtered and ranked on what a page is, not only on what it happens to say.
- A quality score
- Heading structure
- Internal links out
- Language
- Published date
- Author and site
- Page type
- Structured data
- A subject collection
- A safety classification
Raised to date: €0.
Everything described on this page — the index, the fleet reading into it, the workspaces on top of it — was built without a round, a grant or a loan. The funding ledger has six lines and all six read the same.
- Venture capitalno round, no term sheet€0
- Angel investmentno outside shareholders€0
- Grants and subsidiesnone applied for, none received€0
- Loans and creditnothing borrowed€0
- Accelerator fundingno cohort cheque€0
- Revenue reinvested from elsewhereno other business behind it€0
askFinz began on 1 September 2025. Since then the cost has been measured in evenings and weekends rather than in invoices — which is why the platform exists at all, and also why it is smaller than it should be.
Time is the one input that cannot be stretched further. That is the whole argument on this page: the expensive, uncertain part is finished and paid for, and the part that is left is the part money is actually good at.
Four things that are unusual, and all four are checkable.
The index is ours
Most AI products pay a search provider per query and wrap the answer in a model. askFinz reads the web itself and files what it reads, so the rate, the reach and the price are set here rather than by a supplier.
See the index29 workspaces, one bill
Chat, Research, Code, Search, Finance, Projects, Mail, Storage and the rest are one platform with one login — not a bundle of separate products with a shared logo. 13 ship today; the others are in beta or named on the roadmap.
Every workspaceBuilt on a small fleet
The reading is done by machines we run, machines partners run, and machines people lend. Growth is a matter of adding hardware to a pool, not of signing a larger data contract.
How the fleet worksThe numbers are public while they move
Pages read, machines online, queue depth and throughput are on a dashboard anyone can open, including during a bad hour. Diligence on this platform does not depend on us preparing a slide.
Open the dashboardWhat your teams pay for a stack today, and what they pay for one.
Tick the tools your people already sign into. Prices are illustrative list prices per seat per month — put your own in their place and the arithmetic still holds.
- Usage past the plan's allowancetop-ups from $2 / 100K
- Storage past the plan's allowance$0.02/GB a month, $0.015 at volume
- Browser extension$8.19 a month
- Searching the web index$0.75–$3.50 / 1,000
- Selling through the marketplace3.5–5% commission
Subscriptions only · askFinz prices live · stack prices illustrative · not a quote
We read the web ourselves, so we set what reading it costs.
Most AI products pay a search provider for every query and pass the bill on, which puts a floor under their price that they do not control. askFinz has no such floor. These are the published rates, per thousand searches.
per 1,000 searches
Published list rates across comparable providers run from $1.00 to $12.00 per thousand, and each carries its own caveats about what a single call actually returns. The full table, with those caveats.
Three things, and money is the third. That order is not modesty.
The expensive, uncertain engineering is behind us. What askFinz is short of now is use, volume and hands — and two of those three can be given by people who are not writing a cheque at all.
Teams using it daily
Real work, in real organisations, at a scale that finds the edges. Every workspace on the platform gets better from being used against problems we did not imagine, and no amount of engineering substitutes for that.
Run a pilot, or introduce us to a team that should.
Volume through the index
Around fifty thousand searches a month is the point where the index stops being an engineering achievement and starts being a business. The capacity to serve it already exists; what is missing is the demand pointed at it.
Route your search or research workload through it.
Capacity
More machines reading more of the web and re-reading the sources that change, and the room to finish the workspaces still marked beta. Reading is the part that scales with hardware rather than with cleverness, which is why it is where capital moves fastest.
Invest, fund a workstream, or lend capacity.
What the next pound buys.
The engineering that is usually the risk has already been paid for. What is left is the unglamorous half.
- 01DistributionGetting the platform in front of the teams it already fits, which is the thing a small team cannot brute-force.
- 02Compliance and assuranceThe paperwork larger customers need before they can buy: reviews, controls, data-residency options.
- 03Finishing the suiteThe workspaces still marked beta or soon — Data, Charts, Knowledge, Legal, Bank and the rest.
- 04Reading more of the webMore machines and more partner sources, including material nobody is currently paid to index.
Everyone assumes this needs a datacentre. It needed neither.
Compute and storage are the two costs that usually decide an index is impossible for anyone small. Here is what askFinz actually spends on each.
The index was built on a graphics card from 2016.
Not on a frontier cluster, not on reserved cloud capacity, and not in a datacentre. The machines that read the web for askFinz run consumer hardware that was mid-range when it was new and is a decade old now.
- Architecture
- Pascal, 2016
- Compute
- 4.37 TFLOPS
- Memory
- 6 GB GDDR5
- Launch price
- about €438
- Architecture
- Current generation
- Per accelerator
- €35,000–40,000
- Populated 8-GPU server
- €250,000–400,000
- Housed in
- A datacentre
- A Raspberry Pi 4B
- A laptop
- A Mac mini
- A rack server
- x86 or ARM
A reading node is a normal computer, so capacity grows by adding ordinary machines to a pool rather than by buying a rack. How a node works · The infrastructure
A stored page costs 22 KB.
That is the cleaned text and the vectors, with the markup, scripts and navigation furniture thrown away. It is the figure that decides whether an index can keep growing on hardware instead of on a balance sheet — so it is read live from the running system rather than typed here.
About 8.3× more pages on the same card. Not like for like: the reference is a raw archive of fetched pages, while askFinz stores processed passages and their vectors. Part of the gap is that difference rather than engineering.
Drag the capacity up to a drive or a rack and the same arithmetic holds. The full comparison, with the workings · How a page is filed
Some of the most useful help is not money. It is capacity.
If you build models, make hardware, run a datacentre or sell compute, there is a version of this conversation that has nothing to do with a round. What each side brings is below, in that order.
AI labs and model builders
Model access, embedding or inference capacity, and evaluation collaboration.
A corpus that is read first-hand and filed with structure — quality scores, headings, links, dates and subject collections — rather than a scrape flattened into plain text.
Hardware providers
Machines, accelerators or single-board devices for the reading fleet.
A reference workload that runs on x86 and ARM alike, from a Raspberry Pi to a rack server, and reports what it does publicly while it does it.
Datacentre and colocation providers
Rack space, power and transit for reading nodes.
A steady, low-intensity, well-behaved tenant: the load is continuous rather than spiky, and every node is identifiable and rate-limited by design.
Cloud and GPU compute providers
Spare, preemptible or off-peak capacity for reading and embedding bursts.
A workload that is genuinely interruptible. A page that does not get read this hour is read the next, so capacity nobody else can sell without a guarantee is capacity we can use.
Four ways to scale with us. One is brought forward for you.
Tell us which one you are and the matching door comes forward. Nothing is collected and nothing is hidden — the other three stay exactly where they are.
Back the round
What exists today was built by a small team on a small fleet. Capital goes to distribution, compliance, and finishing the workspaces already on the roadmap.
- Data room on request
- Monthly written update
- Board observer seat for leads
Run a pilot
One team, the whole suite, one flat price. You see adoption and the work it replaced, rather than a slide about either.
- One sign-in across every workspace
- Your own material stays yours
- Rollout playbook included
Equip a cohort
Cohort licences for incubators and accelerators. Founders get research, finance, code and project work from day one instead of assembling a stack they cannot yet afford.
- Cohort pricing per intake
- Founder workshops
- Co-branded onboarding
Fund a workstream
Back a defined piece of work — regional language coverage, accessibility, public-interest material that nobody is paid to index. Scope agreed in advance, progress reported against it.
- Milestone reporting
- Open evaluation criteria
- Public summary at close
Small enough to name.
askFinz is built in Eindhoven by Paul Varghese. At this size the platform and the person are the same question, so here is the answer before you have to ask for it.
- FounderPaul Varghese — Eindhoven, Netherlands. Writes the platform, runs the fleet, answers the scale inbox.
- StudiedA double Master's at Eindhoven University of Technology, and a Bachelor's at Delft University of Technology.
- RecordEleven competition wins and placements across three years and several ventures, in cloud, applied AI and industry innovation challenges.
- Already folded inFinRock.ai and BIMDash were both brought into the platform and announced publicly — the finance research layer and the model viewer are not roadmap items.
Nothing on this page needs taking on trust.
Asked before, answered here.
- Is askFinz raising?
- We talk to investors directly rather than through a page. Terms, stage and what is already committed are covered in the data room, which we send on request to funds and angels who ask.
- What does a pilot cost?
- A pilot is one team on the published plans for a fixed period, at one flat price agreed in advance rather than per workspace. The plan prices on this page are the same ones the checkout charges.
- Do you offer cohort licences for accelerators?
- Yes. Cohort pricing is per intake rather than per seat, so an incubator can equip a whole batch of founders at once, with co-branded onboarding and workshops if useful.
- Can a grant or subsidy fund a specific piece of work?
- Yes, and it is the arrangement we prefer with public funders: a defined workstream — regional language coverage, accessibility, public-interest material — with milestones reported against criteria agreed at the start and a public summary at the end.
- How is this different from an AI product built on someone else's search?
- Most AI products pay a search provider per query. askFinz reads the web itself and files what it reads, so coverage, freshness and cost are set here. The index, the machines reading into it and the figures describing both are open on a public dashboard.
Sometimes the useful thing is knowing the right person.
Not every helpful message is a cheque. An introduction to the partner who actually runs the fund, the team inside a company that has this problem today, or the programme manager who knows which call we fit — those move things further than most meetings do.
So this asks one thing plainly: can you decide, or can you reach whoever does? Both are worth telling us, and the answer decides who reads it and how quickly.
- It lands in one mailbox and a person reads it.
- You get a reply — including when the answer is that we are not the right fit.
- If you offered an introduction, we send you something short enough to forward without editing.
Would rather read it than write to us? The pitch deck is online, to page through or to sit back and listen to, and you can take a copy away.
Open the deckWe send a code to your email to open it.
Every claim on this page has a page behind it. Nothing here needs a call first.
If any of this is yours to decide, write to us directly.
One address, read by the people who built it. Say which door you came through and what you would need to see next. If the useful thing you have is a name rather than a cheque, that is worth as much — there is a short form for it.