Skip to content
unravel
DAStart a project
unravel
ContactStart a project

Unravel AI · Copenhagen

Product thinking / Software engineering / Applied AI

We turn AI capability into useful software.

Unravel is an AI software house in Copenhagen. We build custom systems with companies and develop products of our own — combining product judgment, software engineering, and applied AI research.

From possibility to practiceUNRAVEL / SYSTEM STUDY
Layered acrylic, modular hardware, and a visibly routed cable on an AI studio worktable
Business intentWork worth changing
Company contextPeople · tools · data
Working software
UsefulBoundedOwned
We connect model capability to company context, clear boundaries, and an outcome people can own.

Our position

AI capability is becoming abundant. Judgment is not.

The difficult part is no longer making a model respond. It is choosing the right problem, shaping dependable behavior around it, and fitting the result into the way a company already works.

That is where we work: between what a model can do and what a company can safely rely on.

Our products are working arguments

Why these agents

Coding agents changed how software is made. They also exposed the next constraints: understanding large workspaces safely and keeping browser journeys reliable as interfaces move. Khadim and Blop are our answers to those two constraints.

Khadim represented as a contained local workspace of modular tools, context layers, and explicit approval

Khadim

Context that stays close to the work

A local-first AI workspace for coordinating models, tools, repositories, and human approval through an inspectable event stream.

Explore product

Blop

Browser journeys that remain evidence

Agentic browser testing that turns intent into reviewable tests, runs them in real browsers, and keeps failures useful in the repository and CI.

Explore product
Blop represented as three observable browser-test checkpoints with accumulated evidence

How we make decisions

What should survive the demo.

TRACE

Artifacts over dashboards

Code, tests, events, and evidence should remain portable and inspectable. The work matters more than the interface around it.

SCOPE

Bounded action

An agent needs tools and authority, but neither should be ambient or invisible. Access, approvals, and escalation belong in the design.

TRUTH

Honest maturity

We label shipped, alpha, beta, early access, and planned work differently. Trust starts with saying what a system can do today.

Client work

We build with companies — and in public.

For clients, we build AI products, workflow agents, integrations, and the infrastructure around them. Khadim and Blop are our own products and practical proof of the systems we know how to deliver.

  • AI products
  • Workflow agents
  • Integrations
  • Evaluation systems
  • Production infrastructure

Bring us a workflow worth changing.

We will help find the smallest useful system, build it against real work, and leave your team able to understand and own it.

Work with us