Dweve

About Dweve | European AI Company

Dutch AI company building neurosymbolic architectures with explainability, replayable evidence and European deployment. Bootstrapped and independent.

About Dweve

Dweve is a founder-funded Dutch AI company based in Arnhem. It builds neurosymbolic systems with explicit constraints, explainability, replayable evidence and European deployment, while publishing open foundations for inspection.

  • The page introduces five founders and separates company facts from product, research and foundation records.
  • Claims about security, deployment and measured results should be checked against the linked product and Trust records for their stated scope.

Choose the audience that matches your question

The page contains three selectable readings of the same subject.

For consumers

Dweve is a founder-funded Dutch AI company based in Arnhem. This route introduces the five founders and points to product, research, open-foundation and Trust records that explain what the company builds and how to inspect it.

For businesses

Dweve builds neurosymbolic AI systems with European deployment and inspectable records. The company page separates founder and ownership facts from the product, research and foundation records that support a buyer review.

For engineers

Dweve describes an owned path from numerical operations to the workspace, with explicit constraints, replayable evidence and open foundations. Product and Trust records carry the scope for implementation and measured claims.

One platform, one jurisdiction, one auditable record.

Run it managed Fabric, licensed customer infrastructure, or air-gapped. The contracts stay the same across every environment.

Same input, same build, same result. Any run can be reconstructed and re-examined after the fact.

A person keeps the final say on decisions that matter. The system advises, it does not overrule.

Every decision carries a constraint trace. Every agent action runs through policy and is recorded.

All customer data stays within the European Union. No reliance on non-EU infrastructure or control planes.

Deterministic behaviour and replayable records for financial entities and their critical providers.

Isolation, logging, and incident readiness for essential and important entities running the stack.

Explainable by construction with human oversight on decisions that matter. Built for the risk tiers it defines.

The applicable agreement states the processing boundary, transfer conditions, subprocessors, and operational responsibilities.

Founder-funded to date (pre-seed), Dutch, independent

Build neurosymbolic AI that businesses can actually use, understand, and afford. Explainable, efficient, and affordable without the complexity or dependency on foreign technology.

European leadership in AI technology through indigenous innovation, eliminating technological dependencies while advancing sustainable, explainable artificial intelligence.

Binary computation research, constraint-based reasoning, and 2,900+ hardware-optimised operations across the full platform suite: Nexus, Aura, Spindle, Mesh, Loom, Fabric, and Core with enterprise-grade capabilities.

Dweve formally established as a company, bringing a production-ready AI ecosystem for European digital sovereignty to market.

In practice: efficient, explainable, kept in the EU.

In practice: AI that Europe owns, not AI that Europe rents.

Useful AI should not need a data centre or burn energy.

The reasons behind every answer stay visible.

Data and the company stay under European rules.

Five founders, and no one else owning it

If an opportunity violates one of these, we turn it down.

EU-scoped where selected; transfer path declared per posture

Sovereignty is not a marketing badge, it is an architecture decision. A selected customer-operated air-gapped posture can avoid outbound dependency; the public website and managed control plane may use Cloudflare. Provider, transfer and access scope follow the selected deployment and contract. Built for the way Europe regulates AI.

European data stays in the declared boundary where contracted.

You should see how your AI reached a decision before you stake a contract, a patient, or an audit on it. We design explainability into the system through crystallised constraints, not a plausible story reconstructed after the fact. Every decision is recorded for review, subject to the selected deployment and evidence scope.

A higher score on a leaderboard is not a product. Innovation that matters comes from work businesses can actually deploy, understand, and afford. We measure ourselves by problems solved in production, not by points on a chart we tuned to win.

Solving real problems, not chasing benchmarks.

Computational resources are finite, and most AI brute-forces problems that careful design could solve cleanly. We build on binary computation and constraint reasoning that runs on regular CPUs, not warehouses of GPUs. Every cycle we save is a cycle the planet does not need to generate.

When a shop charges you, you get a receipt that shows each item. Dweve builds AI that works the same way. If it tells you something, you can ask to see the reasons, written out plainly.

The computer never has the final word on something that matters to you. It is a helper, not a boss. A real person can look again and change the answer, the way you would ask a second doctor.

Your information stays in a locked drawer here in Europe. It is not sold, not shipped abroad, and not quietly fed into a machine to make money off you.

A lot of AI runs on giant machines that burn enormous amounts of electricity. Dweve runs on ordinary computers and uses far less, the way a small car sips fuel next to a lorry.

No jargon, no surprises. Just fair treatment you can feel.

Arnhem is our primary hub for research, core engineering, and company operations.

A selected customer-operated data plane can keep customer data within a declared European boundary; the public website and managed control plane may use Cloudflare. Provider, transfer and access scope follow the selected posture and contract.

Designed from the ground up for GDPR, the EU AI Act, NIS2, and DORA requirements.

One declared jurisdiction where selected

Data follows the selected European jurisdictional boundary.

Product, engineering, operations, strategy

A patient helper for everyday things, and you stay in charge of every one.

It can draft a note, a reply, or a form in plain language. You read it, change anything you like, and only then send it.

When there is a choice to make, it can lay out the options clearly and, importantly, show you the reasons, so the choice stays yours.

It can look through a pile of papers or messages and pull out the one thing you were after, so you do not have to hunt for it.

"What does this letter from the council mean?"

You can ask it things in ordinary words, the way you would ask a helpful person at a desk, and it answers back plainly.

Not a faceless website. A real company in a real town, run by people you can reach.

Whatever you share stays inside Europe. Nothing is shipped abroad.

A real place in Arnhem, in the Netherlands, with people who work here every day.

Real people run it, and you can talk to them. English, with Dutch a nice extra.

Founder-funded to date, focused on the technology.

Direct investment in technology development by the founders to date; the proposed seed is not yet closed.

Founder capital has gone into technology to date; the proposed seed is not yet closed.

Pre-seed as of August 2026: no dilution or external control plane over the IP.

No quarter-end pressure to ship before it is ready.

Set by engineering and customers to date; the proposed seed is not yet closed.

Core, Loom, Nexus, Spindle, Mesh, Aura, Fabric

Across 76 categories, 12 backend targets

Fourteen foundations, public and replaceable. All Rust except Jacquard.

Every repository lives on github.com/dweve-ai under its stated project licence. FMI uses Apache 2.0; Twin uses Business Source License 1.1. Rust throughout with TypeScript for Jacquard.

Verifiable replayable agent ledger, Ed25519-signed

Dense encoding, 56% fewer tokens than JSON in one measured run

Every figure is measured, not modelled. No estimates rounded up.

14 dtypes and 11 Q-formats, correctly rounded

Value per token versus JSON, same payload

Hot-path policy evaluation, replayable proof machine

2,900+ ops across 76 categories, 12 backend targets

PAP routing, a constraint trace emitted per output

Every step carries identity, version, provenance, and policy context.

Every step is recorded append-only and signed. Any run can be reconstructed exactly, with a proof and audit trail intact.

The same contracts run managed Fabric, licensed customer infrastructure, distributed Mesh, or fully air-gapped, with no rewrite.

Tools and agents run through policy, permission, and audit. Selvedge executes untrusted artefacts under proof and transcript.

A 528-domain-specialist mixture routes via PAP and is constrained by symbolic rules. Lattice evaluates policy in 1.36 ns on the hot path.

Deterministic binary retrieval returns exact matches at recall 1.000, no similarity drift between identical queries.

Kernels run across 76 categories and 12 backend targets on correctly-rounded fixed-point arithmetic, bit-exact where it matters.

Computation is lowered into explicit graph and runtime contracts. Reed produces Merkle-certifiable, content-addressed structure.

Raw sources are gathered, parsed, structured, canonicalised, deduplicated, scored, and given source provenance.

Dweve is a company in the Netherlands, in a town called Arnhem. Five people started it with their own money. No big investors tell them what to do. They build computer helpers that can explain themselves, the way a good neighbour explains why they did something.

Most AI today is like a stranger who gives you an answer but will not tell you how they got it. Dweve makes the other kind. When it gives an answer, it can show you the reason behind it, step by step, in words you can read.

It runs on ordinary computers here in Europe. Your information does not get shipped off to a faraway country. It stays where the rules protect you, and it is not used to train anything behind your back.

Built by people who think you deserve to understand the answer.

Same inputs, same build, same bytes. A sealed run can be replayed exactly on any machine to reconstruct the decision.

Inputs are parsed into content-addressed subtrees, so any fact in a run can be proven without trusting the store.

Each run produces a Merkle root over its events, signed with Ed25519. The signature verifies the whole transcript.