01 / Prospective
Simulation & digital twins
Coherent long-horizon behavior across evolving simulated environments and their real-world counterparts.
Cognitive infrastructure for interactive AI
Tolkun Labs develops Large Cognitive Models to make AI systems reliable over complex & long horizon tasks — remembering what matters, reasoning under uncertainty, working with people, and learning from what happens next.
Emerging from the €1.03M Large Cognitive Models (LaCoMo) research commercialization project, supported by €822K in Business Finland Research to Business funding.
01 / Problem
AI is moving from one-shot generation into long-running interaction. In real operations, systems must preserve context, follow constraints, adapt to change, and recover when things go wrong.
Adding a longer prompt, another tool, or another retry can improve a demonstration. It does not necessarily create a system that people can understand and trust over time.
LLMs provide capability. LCMs enable reliable interaction.
02 / Solution
Preserve what matters without carrying the entire past forward.
Connect goals, constraints, prior decisions, and new evidence.
Record what the system observed, inferred, chose, and did.
Represent uncertainty and communicate it appropriately.
Turn outcomes and failures into measurable system improvement.
Explain the intrinsic decision-making processes in human-readable format.
03 / Use cases
Tolkun Labs is exploring collaborations in settings where AI must maintain context, respect constraints, and remain understandable as situations evolve.
Coherent long-horizon behavior across evolving simulated environments and their real-world counterparts.
Systems that preserve context across incidents and remain inspectable under adversarial pressure.
Agents that plan under uncertainty, act with traceability, and recover deliberately from failure.
Assistants that maintain state across shifts, hand-offs, and interruptions without losing the thread.
Instructive scenarios where reasoning, constraints and after-action review are first-class.
Interfaces that expose evidence, uncertainty and the reasoning behind a suggested course of action.
Listed domains represent areas of prospective research, collaboration and pilot interest. They are not statements of existing customers, completed deployments, or validated performance.
04 / Our thesis
Human intelligence does remarkable work under limits. It selectively remembers, redirects attention, revises interpretations, and learns through action.
Tolkun Labs applies insights from computational cognitive science to interactive AI — not to copy the human mind, but to investigate systems that can operate coherently under limited context, changing conditions, and real-world consequences.
The question is not only whether an AI system can perform a task once. It is whether it can behave reliably over time.
Further reading
A longer essay on why interactive AI needs cognitive structure around foundation models — memory, action, uncertainty, and learning.
05 / Why Tolkun?
Tolkun comes from the Finnish word tolkku — a sense of order, reason, intelligibility.
Reason.Order.Coherence.Intelligibility.Sense.
It is what a situation needs when it has become confused or excessive — when it needs to start making sense again. That is our ambition for interactive AI.
Collaborate
Building something that has to remember, adapt, and stay accountable? We should talk.