We build AI that is safe, affordable and open to everyone.
We build AI that people and companies can actually rely on — software that checks its own answers, keeps your data on your side, and costs a fraction of what AI usually costs to run. The same engineering serves one person on a laptop and a team with rules to answer to.
- Company
- Lotus Mind
- Type
- AI systems company
- Founded
- 2025
- Based in
- Dublin, Ireland
- We serve
- People and companies
- Products
- 6 built in-house
AI should be safe, affordable, and open to everyone.
Today the best AI goes to whoever can pay the most. Large companies get private, governed, carefully engineered systems. Everyone else gets a chat box, a monthly subscription, and a quiet warning not to trust the output.
We do not think that split has to exist. It exists because most AI software is built to burn as many model calls as possible, and to keep your data somewhere the vendor can see it. Both of those are choices, not requirements.
Open to everyone also has to mean you can leave. Our tools run on your own machine, what they remember is a file you can read, and your work stays in your own hands rather than ours.
We build the opposite: AI that a person, a small business and a bank can each trust — and each afford.
Make AI safe enough to rely on, cheap enough to use, and simple enough for anyone.
Safe
An AI that works on your behalf has to be honest about what it does not know, and it has to ask before doing anything you cannot undo. We treat that as a design rule rather than a setting. And your work stays yours — on your own machine, or inside your own systems.
Low cost
Using AI should not need an enterprise budget. We build so that ordinary software does most of the work and the model is called only where real judgement is needed. Much of what we make costs nothing at all to run.
Simple
You should not need training, a setup project or a consultant before it is any use. You say what you want in ordinary words and the software does the technical part. If something needs a manual before it helps you, we have not finished it.
For everyone
One person, a small business, or a large company — the same standard of safety and the same quality of engineering. We do not ship a weaker, less careful version for the people who cannot pay much.
One person, a small business, or a whole company
Most AI companies pick one of these and ignore the other two. The safety and cost work underneath is the same either way, so we did not have to choose.
One person, a laptop, and a life to get on with
A companion that remembers you and belongs to you. An academy that teaches a child how AI actually works. A file too big for a spreadsheet, opened and answered on your own machine. Free, or close to it, and private by default.
- Project Nobi
- Aika
- Datalic Sheets
- JenOS
Agents that need somewhere safe to work
Run your coding agents in isolation and review everything before it lands. Put a governed, verified data layer behind your own product with an SDK, an MCP server and drop-in widgets.
- Lopy
- Datalic SDK
- Datalic MCP
Real data, real rules, and an audit trail
Governed AI over the databases you already run, deployed inside your own network, where the cloud half has no way to reach your data at all.
- Datalic
- JenOS
- Datalic Sheets
Five things, one company
Not an agency that connects APIs, and not only a product company. We advise, we build, and we own the software it is built on — which is what lets us serve a single person and a large company from the same foundation.
We design and ship complete production systems — agents, retrieval, evaluation and the plumbing around them. Not prototypes handed over as a slide deck.
Opportunity assessment, workflow analysis, data readiness and an adoption roadmap. We start by finding the problem that is worth solving.
Internal tools, assistants and workflow automation built on your own systems, with a person kept in the approval loop.
Safe, governed access to data across 32 database engines and the files on your own machine — the layer that decides what an AI is allowed to see and do.
Our own software — some of it free to run on your own laptop, some of it sold to companies. Client projects are delivered on the same foundation, which is why they take weeks rather than quarters.
Why Lotus Mind
The name captures what we ask of every product we build and ship.
A lotus does not grow in clean water. It needs the mud. And it is not a rare flower kept behind glass — it grows in ordinary ponds, in ordinary places. That is how we build: for the messiness of real life, not the neatness of demo data; for everyone, not just those who can pay the most.
And nothing stays on the leaf. Water rolls off without leaving a trace. Your work passes through our products, but we keep none of it.
The “mind” is simpler. A clear mind is not one that tries to hold everything. It is one that knows the difference between what it knows and what it does not. Our software respects that boundary. It does not invent what is missing. It asks before taking actions you cannot undo. And it is designed to be useful without requiring a manual.
That is Lotus Mind: built for the real world, clear about what it knows, respectful of what belongs to you, and simple enough to just work.
What we will not trade away
These are constraints, not slogans. Each one rules something out that would otherwise be easier to build.
A checked answer, or none
A confident wrong answer is worse than an honest "I don't know." Every system we build is able to abstain, and says so when it does.
Your data never leaves your side
We would rather redesign a product than ask you to hand over your data. In practice that means files are read in your own browser, and the cloud half of our platform is built with no way to reach a customer database at all.
Ask before acting
AI can do any amount of work. It should not send, publish, pay or delete without a person saying yes.
A model call you did not need is a bug
Models are for judgement, not for work that ordinary code does correctly every time. Every call we avoid is money the person using it keeps, and one less thing that can behave differently tomorrow.
Plain language, always
If a feature cannot be explained in two short sentences, it is not finished. That applies to our documentation as much as our products.
Built by the people who run it
We use every product here in our own daily work. The bugs find us before they find you.
Four steps, in this order
Most AI projects fail at step two, so we do step two first.
Find the real problem
A short discovery conversation, then a written assessment of where AI helps and where it does not. Often the honest answer removes work rather than adding it.
Prove it on your data
A small working prototype on your real data, not sample data. This is where most AI projects quietly fail, so we do it first.
Build the system
Delivered on top of our own platform, so the foundation — connectors, governance, verification — already exists and is already tested.
Run it with you
Hosting, monitoring, model updates and training. You can also take the whole thing in-house; it is designed to be self-hosted.
How we got here
The data layer
Datalic starts: schema introspection, query execution and the first connectors across the SQL engines.
Datalic Studio
The AI database client — browse objects, run queries, edit results, with an AI analyst beside you. Web, desktop and tablet.
Governance and verification
The verify-or-abstain loop, connection access control, workspace isolation, the semantic layer, and a governed MCP endpoint for agents.
The platform split
Cloud brain and local SDK separated, so query generation runs in the cloud while credentials and execution stay inside the customer's network.
Beyond data
Lopy and JenOS put agents to work under review. Project Nobi and Aika take the same rules to people and families — memory that belongs to you, and an AI that admits when it is wrong.
What we can take on
Applied AI engineering
Agent loops, tool design, retrieval and evaluation — built to be measured rather than demonstrated.
- Agent and tool design
- Retrieval over SQL and NoSQL
- Answer verification
- Model-neutral architecture
Data platform engineering
Connectors, schema introspection, a semantic layer and governance across the databases a real company runs.
- 32 database connectors
- Semantic contracts
- Row and column policy
- Audit and isolation
Product engineering
Desktop, web, tablet and embedded interfaces, designed alongside the systems behind them.
- React and TypeScript
- Electron and Capacitor
- Chat and voice interfaces
- Embeddable widgets
Platform and deployment
The same product running on a laptop, self-hosted in your Docker, or in a managed cloud — chosen at runtime, not at build time.
- Docker self-host
- Per-workspace storage
- Multi-tenant routing
- MCP endpoints
Have a problem worth solving properly?
Tell us what you are trying to put AI in front of. We will say plainly whether we can help, and if we are not the right fit we will say that too.