Download the press kit (ZIP, logos and boilerplate)

Download the press kit (ZIP, logos and boilerplate)

About Verkko

Verkko is an applied research lab that designs and builds thinking systems that learn like brains do: constantly, without forgetting, and without wasting huge amounts of power. It is part of the Verkko Robotics Research Lab in Europe. Verkko's mission is embodied living intelligence: AI that learns continually from use, keeps what it learns, and can act in the world, accountably.

Boilerplate

Verkko is an applied research lab building embodied living intelligence for businesses, people and machines. Its products, Verkko Nexus, Verkko Halo and Verkko Kinetic, run on Voltaic, a from-scratch spiking neural network that learns continually at inference time and forgets far less than conventional models. Verkko is part of the Verkko Robotics Research Lab in Europe. Learn more at verkko.ai.

Products

  • Verkko Nexus: the AI operating layer for organisations. It connects the tools teams already use and puts governed agents to work across them, with human approval before anything sensitive goes out.
  • Verkko Halo: a personal intelligence that stays with you. A morning brief, a memory you can ask and routines that draft before you ask.
  • Verkko Kinetic: the physical AI of Verkko Robotics. Real robots that run the same spiking core in the field: they learn skills by demonstration, keep what they have mastered and run on very little power.

Technology

  • Voltaic: a from-scratch spiking neural network built from binary spikes, LIF neurons and STDP attention. It learns at inference time and forgets far less than conventional models.
  • MemoryForge and ContextFlow: a persistent memory that keeps what's important and assembles the right context for each task.
  • Morpheus: a knowledge graph that learns from how teams actually search.
  • Baton: an vendor-neutral protocol for governed AI agents, where every action is scoped, recorded and attributed.

Key facts

  • 0.04% forgetting on CORe50-NI with HULL, down from 1.20% for a standard model.
  • 4.2× more knowledge retained than dense 7B baselines on Split ImageNet-1k, and 2.6× on Split CIFAR-100, with no replay buffer.
  • Runs on a single GPU at 355M to 7B parameters.
  • 78 ready-made use cases across 10 departments and 6 industries.

Brand assets

Please don't alter, recolour or distort the logos. Use the dark versions on light backgrounds and the light versions on dark backgrounds.

Media contact

For interviews, briefings and image requests: press@verkko.ai