Runheim lets anyone in a company build AI agents that work in the systems it already uses, in every part of the business. We build on top of Claude, ChatGPT and Copilot: the AI can ask; only Runheim can act.
We built the AI agent platform at a large financial institution. It went company-wide in under 2.5 months, and more than 550 of its 700 employees use it today.
We ran critical systems in finance and offshore, and we know where they break. An agent is only useful when it can reach those systems safely, so that is the layer we build: we connect to each company’s systems, read and write, such as SimCorp Dimension’s APIs, and set them up so agents can use them safely. We take each company through how it works, so its own people set the policies.
We have worked side by side at the same institution since 2022, including seven months when Magnus was seconded to Johan’s security team. Johan has the final call on product and architecture, Magnus on sales, customers and deployment, and we both write code every day.
We argue briefly and decide fast. Adding automations and dashboards to the platform looked like scope creep, so we debated it, shipped it quickly, and had it before Claude and ChatGPT did.
Outside work we made 17,000+ GitHub contributions to our own projects in the past year, and found a flaw in Claude Code skills that was fixed after our report. Runheim is the two of us, as equal partners, and you talk to us directly.
I have worked in security since 2018, starting as a cyber security analyst at mnemonic, one of Norway’s largest security firms. As a Telenor intern in 2020 I built a GitOps workflow on Kubernetes and presented it to Telenor Norway’s board and the Telenor Group CEO.
At a large financial institution since 2021 I have carried the security deliveries the company depends on, from cloud and infrastructure to the controls on top. I benchmarked which AI models can be trusted with security work and delivered the agentic workflows for cyber threat handling. That showed me exactly where agents break: the line between what an agent proposes and what a real system lets it do.
I created Runheim and wrote most of its core in about four weeks: a policy gateway that checks every action against the company’s rules, a key broker so agents never hold credentials, approvals bound to one exact action, and an audit trail. Outside work I code just as much, from trading systems to home automation. MSc Information Security, NTNU.
With Johan Selnes, I led the six-person pod that built the AI agent factory at a large financial institution: first users in three weeks, company-wide in under 2.5 months, and more than 550 of its 700 employees now use agents. It worked because anyone could use it and it was wired correctly into real systems. Most companies stall on that second part, which is why we started Runheim.
In the Royal Norwegian Navy I was chief engineering officer on a 60-knot Skjold-class corvette, responsible for its four gas turbines and an engineering crew of six. I joined the same institution as an analyst in 2022; four years later I lead the team running the core systems every trade there goes through.
Working evenings and nights, Johan and I built Runheim in six weeks by directing 100+ coding agents through Gram, the orchestration tool we wrote. MSc Industrial Economics, NMBU; BSc, Royal Norwegian Naval Academy.