Skip to content
Applied AI, product engineering, and notes from the workLearn more
unravel
DAContactStart a project
unravel
ContactStart a project

AI software, built around your business.

Unravel is a Copenhagen AI software house. We design and build dependable agents, intelligent workflows, and custom software for companies, while developing products of our own.

Discuss a project
WORK TO IMPROVE

Turn our manual case intake into a dependable AI product

CHOOSE A SYSTEM
01
Product UI
02
Bounded agent
03
Evaluation
READYDELIVERABLE

A working product your team can inspect and own

  1. 01Map decisions and source data✓
  2. 02Build the narrow end-to-end flow✓
  3. 03Evaluate against real cases✓
Strategy · Product engineering · AI systems · DeploymentSee our work ↓

From a useful idea to working software.

We join business context with hands-on engineering. The result is not another AI presentation, but a system your team can use, inspect, and own.

BUILD01

AI product development

New products and customer experiences built around language models, multimodal systems, retrieval, and agentic interaction.

AUTOMATE02

Workflow automation

Bounded agents that connect your tools, data, and approvals to remove repetitive work without removing accountability.

INTEGRATE03

AI integration

Production integrations across internal software, APIs, Microsoft and cloud platforms, with security and observability designed in.

ADVISE04

Technical direction

Architecture, prototypes, evaluation, and delivery plans for teams that need to turn AI ambition into an executable roadmap.

Useful in a demo is not enough.

We define what good looks like before scaling a system. Every delivery is tested against the work it must perform, the boundaries it must respect, and the people who need to trust it.

Evaluation reportExample structure
Input

A representative production task

01
Outcome quality

Does the result solve the actual task against representative cases?

02
Boundaries

Does the system stop, escalate, and ask for approval at the right moments?

03
Operations

Can your team inspect failures, costs, sources, and decisions in production?

We build our own products too.

Khadim and Blop are where we test our beliefs in public: agents should act on real work, expose what they do, and leave durable artifacts behind.

Khadim Electron home screen with projects, model controls and AI composer
KHADIM / DESKTOP

A local-first AI workspace

Khadim coordinates models, tools, projects, apps, and human approval from one desktop workspace.

ExploreK
Current Blop web interface showing a failed run becoming a cluster and verified fix pull request
BLOP / WEB

QA tests owned as code

Describe journeys in plain English. Blop runs them in real browsers, clusters failures, and keeps the tests reviewable in your repository and CI.

ExploreB

Product practice

Products are proof, not the limit of our work.

The same engineering behind our products is available to client teams: agent architecture, local and cloud runtimes, browser automation, MCP, evaluation, and interfaces people can trust.

Built for the systems your business already depends on.

Production AI is an operating model, not a model call. We design ownership, access, review, and observability into the software from the beginning.

01

Your data stays bounded

Access follows the least privilege needed for the workflow, across local, private cloud, or managed infrastructure.

02

People keep authority

Approvals, exceptions, and escalation paths are explicit parts of the product rather than afterthoughts.

03

Behavior stays inspectable

Sources, tool calls, costs, failures, and evaluations are visible enough to operate and improve.

04

Your team can own it

We deliver understandable architecture, durable artifacts, and a clear path beyond the first release.

A small senior team, close to the problem.

We start with the workflow and its constraints, ship a narrow end-to-end version, measure it against real work, and expand only where the evidence supports it.

  1. 01

    Frame

    Map the work, users, risks, and the outcome worth changing.

  2. 02

    Prove

    Build a working vertical slice against real systems and data.

  3. 03

    Ship

    Harden the product, integrate it, and leave your team in control.

Applied research, with a route to production.

We track new model capabilities, agent runtimes, evaluation methods, and interaction patterns by building with them. Research matters when it changes what can be shipped responsibly.

Read our field notes

What should AI make easier in your company?

Bring us the workflow, bottleneck, or product idea. We will help identify the smallest useful system and build it with you.

Start a conversationC