What we do

We offer four things, and each one is something we run ourselves — which is the only reason we are willing to sell it.

How we work — and what we will not take

We take few engagements and we say so up front. Ľubomír carries a long-term delivery commitment in biometrics and security technology; Digital Symbio takes on work that can be done properly alongside it. So we take on less work than we could. What you get out of that is a supplier who does not need your contract — and who will tell you when you do not need us.

What we do well

  • Scoped engagements of two to six weeks, with a defined end.
  • An audit that ends in a written verdict you can act on without us.
  • An agent pilot on one team's real workflow, carried all the way into production.
  • An application with a defined edge, shipped and handed over.

What we will not take

  • Staff augmentation or a body on a seat.
  • On-call, 24/7 response, or anything with a four-hour SLA.
  • Biometrics — face or fingerprint recognition — or work competing with our clients' business.
  • Anything we would have to learn on your budget.

Infrastructure agentization

The problem

Your developers are already pasting code into a chat window. You just cannot see which ones, or what went with it. Meanwhile the boring operational work — the checks, the escalations, the 3 a.m. triage — still needs a human, because you do not yet trust any of those tools enough to let them act alone.

What we do

We put LiteLLM in as the only permitted exit for model calls — one place where "what may go where" is decided. Each application gets its own key, scoped to specific models; never the provider's raw key, which can be copied and walked out of the building. Budgets per project, and a trail of what went where. Sensitive routes are pinned in that config to an Ollama on your own GPU — including the embeddings behind document search, which are therefore computed on your own hardware. The boundary is a line of configuration: you can read it, change it, and audit it the same way you audit any other rule on your network. Then agents on top: read freely, write only on confirmation, every capability declared in one registry rather than scattered through prompts.

How we know

We run exactly this. Twenty model routes behind one gateway, per-project keys, agents on cron making real decisions unattended, and a Signal-based agent that transcribes a voice note and acts on it. A shim we wrote in front of the Anthropic API keeps billing on a subscription rather than per token — the one model that did bill per token asked for $27.52 in a month and left the gateway. Before any of it went live, we ran a four-expert adversarial review of our own design and it found a live credential-exfiltration path in our own tooling. We found it. Not an incident.

Typical first step: a two-week audit with a written verdict on what your teams are actually sending where.

The software you are missing

The problem

You have a process held together by a spreadsheet, an email thread, and one person who remembers how the whole thing goes. Off-the-shelf software does not cover it, and a large vendor will build it at a price that makes the whole thing not worth doing. So it stays as it is — and the spreadsheet keeps growing.

What we do

We build one thing with a defined edge. No framework tower: an application that will run on shared hosting, a database you can back up yourself, a deploy that is one script. Where it makes sense, a model does the boring part — recognises the photo, looks up the price, writes the first draft — and a person confirms it. You end up with the code and a way to run it without us. Renting your own software back to you is not our business model.

How we know

Our own applications, in production. Accounting that pulls its own bank statements and sends payments on a single signature. A PWA for thirty-one people at a festival with no signal — no database, state in files, so there is less to break. A bilingual Laravel app serving two domains from one codebase on shared hosting. A tool that turns a photo into a classified ad: the model identifies the item, researches the market price and writes the Slovak listing, a person confirms, the app posts it. And a health backend on PostgreSQL with roughly as much test code as production code, where every warning fails the build. Nobody commissioned any of them. We built them for ourselves and we have to live with them, and that is the only reason we can also tell you what we would do differently today.

See our applications in the Lab →

AI enablement for your teams

The problem

Half your people are quietly using AI and getting good at it. The other half are afraid of it, and a few are afraid it will take their job. A vendor workshop about prompt engineering will not move either group.

What we do

A day on site, on your real workflows, aimed at agentic tasks — work that runs without a human waiting on it. We start from what your teams already do badly with AI, not from a curriculum. And we are honest about what AI is bad at, which is the part that earns the room.

How we know

This is not really a technical problem — it is facilitation. Ľubomír has spent years facilitating communities and is completing Community Building facilitator training, certified by the end of 2026. Walking into a room of people who think a machine is coming for their work, and leaving with them curious instead of defensive, is a different skill from configuring a gateway. We have both.

Home automation

The problem

Smart home products are sold as convenience and delivered as a drawer of apps that stop talking to each other the moment a vendor changes their API.

What we do

One system on Home Assistant, deployed from git through its REST and WebSocket APIs, so every change has a commit and a way back. Presence-aware lighting, EV charging timed to when you actually leave, heating that follows the season without being told, alerts that matter and silence that is real.

How we know

A hundred automations run our own premises across twenty-one rooms. There are more than a hundred devices, most of them speaking Zigbee through a coordinator that sits in our own cupboard. That includes dual-tariff metering: the car charging pauses itself during peak hours and resumes off-peak.

More about home automation →

What it costs

Ballpark figures, not a rate card — the real number depends on what we find. But you should know the order of magnitude before you write to us, and we would rather disqualify each other early than after three meetings.

First hour Free

You describe the problem. We tell you straight whether we can help, roughly what it would take — or that you do not need us. No deck.

AI audit from €4,500

Two weeks. A written verdict: what your teams are actually sending where, which of it is a real risk, and what can be closed within a month. Yours to act on with or without us.

Agent pilot from €10,000

One real workflow of one team, from nothing to running in production. Scoped from the audit, so nobody is guessing.

Training day €1,500 / day

Fixed price, on site, your workflows. For a company of a couple of hundred people this is usually two or three days, split by team.

Prices exclude VAT. We are a VAT payer (SK2120641875).

First hour free