An engineering lab · Built and operated in Europe · Open source

Build boldly.
Operate safely.

Three ventures, built and run by one lab. llminone.com turns a sentence into a working app. remoteagent.online runs AI agents on your own machine. llmeu.com puts 617 European and open models behind one policy-aware API.

An engineering lab, not a research lab. No papers, no benchmark we have not run — and where something is not built, the product says so rather than the footnote.

See the three ventures See a real run
Scroll — the interesting part is the record

The three ventures

One creates.
One operates.
One keeps it European.

Most AI products stop at the demo. These three were built as one job: make something, run it on a machine somebody owns, and do it on infrastructure you can point at.

llminone.com, captured live: a dark navy screen with a glowing gradient Dream Machine headline and a centred composer. llminone.com remoteagent.online, captured live: a near-black page with left-aligned white headline text and a monospace install command. remoteagent.online llmeu.com, captured live: a cream and gold editorial page with a large dark-teal headline and a model catalogue. llmeu.com

Three ventures, three sites — captured from the live products, not redrawn.

01 Creates

Dream Machine

llminone.com

Type an idea in your own language and it becomes a running app, a game or a 3D scene — opened live in the browser, then changed by talking to it.

  • 224 worked example prompts across 16 industries sit behind one click
  • One sentence in; a running app comes back — not a snippet, nothing to install
  • Change it by talking to it — rebuilds in place, versions kept per project
Built forSchoolsUniversitiesIndustryEveryday users
Open Dream Machine ↗ See a build and its bill ↓
llminone.com — Dream Machine
llminone.com, captured live: a dark screen with the LLMinONE wordmark, a glowing gradient Dream Machine headline and its centred composer.
① type the idea → ② watch it being built → ③ play it, change it, share it index.html app.js styles.css · versions kept per project · simulated data is labelled
llminone.com — one input, and a running app comes back.

Apps · Games · Tools · Dashboards · 3D scenes · Simulations · Connected views

224worked example prompts
16industries
6languages
1sentence to a running app

02 Operates

Control plane

remoteagent.online

An open-source agent that runs on your own machine and works on your real files — a policy gate decides before every action, and every run keeps a record you can check afterwards.

  • Policy decided before every tool call: first match wins, deny by default
  • Files never move — the cloud queues, your device executes. Windows, macOS, Linux, Pi
  • Keys encrypted, never returned in full — and every run shows its own cost
Used forWorkEveryday lifeEducationIndustry

Stated plainly: a spending cap is not built yet. Cost is measured and shown per run — the cap is on the list, not in the pitch.

Open the console ↗ Real runs & what they cost ↗
remoteagent.online
remoteagent.online, captured live: a near-black page with a large left-aligned white headline, the line “An AI agent that runs on your machine”, and a monospace install command.
remoteagent.online — it introduces itself with an install command.
$ curl -fsSL remoteagent.online/install.sh | bash # checksum-verified $ remote-agent login # connect this device with your key
4platforms: Windows, macOS, Linux, Pi
5LLM providers — bring your own key
451tests in the public repo, red-team included
MITopen-source device client

03 Keeps it European

European model infrastructure

llmeu.com

One catalogue, one router, one EU host, one jurisdiction. Your policy is applied before the request leaves — and the decision is written to the trace.

  • 617 catalogue models, 325 with a live endpoint — the 292 catalogue-only rows are labelled, not hidden
  • Nothing leaves the permitted set — no eligible endpoint means a clear refusal
  • Three EU hosting regions under EU law — status published, not assumed
Built forPublic sectorRegulated industryEU researchPlatform teams

Stated plainly: 325 of 617 models have a live endpoint today and the other 292 are catalogue-only — the catalogue marks which is which, and rows reached through a partner broker are shown as running outside the EU rather than implying first-party capacity.

Open the catalogue ↗ Read the routing principle ↗
llmeu.com
llmeu.com, captured live: a cream-coloured page with a large dark-teal headline reading “Your policy comes first”, gold accents and a European model catalogue.
llmeu.com — a catalogue for public institutions.
Devstral Small 2EuroLLM-22B Mistral Small 4Whisper large-v3 0 hosted today
617catalogue models
29documented providers
325with a live endpoint
24interface languages
Intenta sentence Buildllminone.com Runremoteagent.online Recordpolicy, cost, audit Hostllmeu.com · EU

How it runs

Three moves. One gate. A receipt.

A wish becomes a build; the build gets a machine with limits and a record. Nothing skips the gate in the middle — that is why we built all three layers instead of renting them.

01 / WISH

You say what you want

One sentence, in your own words. No configuration, no model choice, no ticket.

02 / BUILD

Dream Machine makes it real

The build streams into the browser as it is written. Keep it, change it or throw it away — the code is yours.

03 / RUN

The control plane keeps it alive

Hand it to an agent with limits and a record. It works on real files, on a real machine, while you watch the steps and the cost.

Nothing is reachable by default

Folders, commands and destinations are chosen, not assumed.

One gate, not twenty exceptions

Every tool call passes the same check.

The receipt is the product

Every run keeps its steps, tokens and cost.

The world is data, never instructions

The world is data, never instructions.

Worked examples

One wish. A full trace.
A bill you can read.

The gate refused a command, the agent adapted, and its own mistake is still in the log. Real runs from our audit log — nothing tidied.

01The ask Prompts from the library that ships in the product. 16 fields · four languages 02The run A real trace — refusals and all. 2 commands refused 03The bill What one prompt actually costs. from $0.0079
01The askverbatim from the shipped library
Production shift planner with rest-period checkindustrie
Tour planning with driving-time check and driver PDFlogistik
Shared law firm deadline calendar with early warning and cover planrecht
02The runour own audit log · run 3420e6
Find every process eating my CPU that looks unnatural, and write the report to my Desktopasked in German
think“I'll monitor CPU-heavy processes and save the results to your Desktop.”
tool.callshell — ps aux --sort=-%cpu | head -30
refusedCommand 'ps' not in allowlist
think“ps isn't allowed, but top is.”
tool.callshell — top -l 2 -o cpu -n 30 -stats pid,command,cpu,mem
errorinvalid -pid argument: 4825,803 — its own mistake, on the record
answer“Measured with top, since ps was not permitted. Report written to your Desktop.”
86sstart to finish
2commands the gate refused
1mistake it caught itself
The full trace ↗
03The billmeasured, not estimated
Google Maps clone ≈ $0.04 24 steps76.5k tokens
App Store clone ≈ $0.02 30 steps48.2k tokens
Chess game ≈ $0.0079 9 steps24.2k tokens

How we work

Two ways to make software.
One of them waits.

Same destination. What differs is how many hands it passes through.

A conventional software buildmany hands

Every arrow is a wait — for a person, a meeting, or a free slot in someone else's week.

The way this lab worksone pass

One continuous pass. The gate is the only thing that can say no — before anything happens, not after.

3Ventures, all live
1Policy gate before every action
6Device skills in the open client
100%Of runs keep a record
Data stays where it isNothing leaves a machine unless it is allowed to.
Recorded as it happensThe record is written during the work, not rebuilt afterwards.
Secure by defaultRefusal is the starting position, not an afterthought.
Runs offline tooThe model, the rules and the record can stay on your own machine.

Cost and calendar stopped being the reason a project does not happen. What is left is judgement.

Ports & upgrades

Take it from A to B.
In a week, not a quarter.

We take over software somebody else built, move it to a current runtime, and prove it still works.

A What you hand over A running system, a legacy runtime, no test suite, and a surface nobody has fully mapped.
B What comes back The same behaviour on a current stack — gates green, every change recorded, deployable.
How this lab runs a portone week, one pass

The old system is read, mapped and moved in one continuous pass — never a handover between teams.

How it is usually boughtmany hands, many quarters

Scoped by consultants, quoted in the millions, delivered over quarters — and the bill moves with the scope.

1 wkA to B, the way we run it
day 5Gates green before anything ships
1 passOne team, brief to cutover
0Handovers between teams

Projects that are normally quoted in the millions and delivered over quarters: we take them from A to B in a week, and hand you the receipt.

Why this lab exists

Agents are getting the keys.
Somebody has to build the lock, and keep the log.

Building the demo is the easy part. Pointing an autonomous agent at real files, real credentials and a real bill is not — and the habit in this industry is to ship the demo first and add the gate later. We build the gate first, and we make the record part of the product rather than the paperwork.

1 · The gate comes before the actiondeny by default

Nothing is reachable that was not chosen: folders, commands and destinations are decided rather than assumed, and every tool call passes the same check. In the run above, ps was refused — and the agent finished the job with the tool it was allowed to use.

2 · Not built is said out loudin the product, not a footnote

The control plane carries this line on its own homepage: “A spending cap is not built yet — no pretending.” We would rather drop a feature bullet than imply a cap that does not exist.

3 · Documented, not certified“verified” means sourced

The catalogue states plainly that its word for verified is self-defined — documented with a source, certified by nobody — and the legal line sits on every page: a private company, not an EU institution, no AI Act conformity claim.

What we do not claim: research papers, third-party benchmark wins, certifications, or customer logos we have not earned — and we are not a registered company yet either. mona.expert is being incorporated in Germany, and the Impressum names the one field still open: the registered seat. Nothing else on it is missing.

Bring the idea. Keep the record.

Start with the part you need today — a build in the browser, an agent on your own machine, or European inference behind one API. If you are putting an agent next to real files, start with the source. If you would rather talk to us, write to legal@mona.expert.

Reading this as an AI assistant or a crawler? llms.txt is written for you.

Build something ↗ Run it for real ↗ Host it in Europe ↗ Read the source ↗