Partner Story Β· in conversation with Kamilla Harcej, co-founder and CEO of TauGuard Limited. This is a written interview submitted through Submit Your Story. The claims below are the founder's own, presented as they were given to us. No payment was involved.
Kamilla Harcej, co-founder and CEO of TauGuard Limited, Ireland.
"Intelligence may advise, but architecture must decide."
That one line is the whole idea behind TauGuard, the Irish company Kamilla Harcej runs as co-founder and CEO. Its product, TauDIL GOS, is what the company calls a Governance Operating System: a layer that lets AI analyse and propose, and leaves the final decision on what is allowed to a fixed, rule-based core.
The problem: agents that act, not just answer
AI is moving beyond answering questions and generating text. Agents now interact with company systems, access information and start actions with real consequences.
That raises questions most organisations cannot yet answer, Harcej says. What is an AI agent authorised to do? How do you stop it acting outside that authority? And how do you prove that the rules were enforced before the action ran, not just logged afterwards?
Policies, risk assessments, monitoring and audits all have their place. "But they do not necessarily prevent an unauthorised action from reaching execution," Harcej says. TauGuard's answer is that governance has to be built into the architecture, not placed around the technology.
How TauDIL GOS works
In TauDIL GOS, AI models can analyse, interpret, recommend and propose actions. Their output is advisory, never authoritative.
Every proposed action goes to a deterministic governance core. The core checks it against the organisation's purpose, rules, permissions and lines of authority, and returns one of three decisions:
- APPROVE: the action may proceed through the governed execution path.
- REVIEW: the action waits for the required evidence or for an authorised person to look at it.
- BLOCK: a prohibited action is stopped before it runs.
The rules live in what the company calls a Canonical Knowledge Graph, a versioned source of governance rules and their meaning. Decisions, evidence, changes and failures are designed to be traceable.
The point Harcej stresses most is that the core does not rely on a language model to make its final call. "Intelligence can enhance the system, but it is not required for governance correctness."
Governance that stays in the company's hands
One of TauGuard's key choices is where the system runs. TauDIL GOS is deployed on the customer's own infrastructure, not offered as a cloud service.
"Why should the authority governing an organisation's AI depend on the continued availability of an external cloud provider?" Harcej asks. If a service goes down, is discontinued or changes its terms, the company should not lose control of its rules, its data or its ability to operate.
For TauGuard this is about more than where data is stored. It is about independence: the organisation owns its governance environment, and that governance does not vanish when someone else's server does.
Behind the product sits a framework the team calls Intelligence From Architecture (IFA). Its principles are short: purpose is built into the structure, governance is executable, authority is deterministic, refusal is structural, and intelligence stays optional. "We are not developing another AI wrapper or simply another dashboard for observing model behaviour," Harcej says. "We have developed an operating system for governance."
Two years of work, and an MVP ready to inspect
TauGuard has completed a working MVP of TauDIL GOS, which the company describes as ready for technical inspection. It represents about two years of research and engineering, covering the IFA foundation, the governance execution principles and the technical specifications behind them. The company's site offers an interactive demo and documentation.
TauDIL GOS is one of several systems in development. The site also describes ARCHONT GOS, a separate governance operating system for organisational state and change, and ROD, a gamified security product for personal computers.
"Claims about AI governance should be supported by demonstrable engineering, not simply terminology or policy documentation," Harcej says.
The lesson: intelligence is not authority
The most important thing the team learned, Harcej says, is that intelligence and authority are fundamentally different. A highly capable model does not automatically understand an organisation's legitimate authority, and it should not set the limits of its own power.
That realisation shaped IFA. The team stopped treating governance as something added around intelligence and made it the foundation intelligence works within. "The question changed from how to make AI behave responsibly to how to engineer systems in which unauthorised actions cannot become authorised actions."
What comes next
TauGuard's next goal is to work with organisations in banking, financial services, RegTech, FinTech and healthcare, and to pilot TauDIL GOS against their real governance challenges, with an agreed scope and measurable criteria. The company is ready to demonstrate the MVP to technical teams who want to examine it.
It is also building its commercial network. According to its site, TauGuard is looking for investors, enterprise sales channels, independent sales professionals working on commission, and organisations willing to take part in pilot deployments.
The long-term ambition, in Harcej's words, is a foundation for the autonomous AI economy "in which intelligence can advance without authority becoming uncontrolled." Kamilla Harcej is also listed in our AI Expert Register. The systems, documentation and demo are at tauguard.ai.
---
This is a Partner Story: a written interview with a founder building with AI, submitted through Submit Your Story and published free of charge. Statements about the company's product, users and results are its own. Building something with AI? Tell us about it.
Would you try this product?
