About us

Accessible to everyone

Our first and most important goal is to make Agent Zero AI accessible to ordinary tech enthusiasts who don't have the time or resources to host their own servers. Many want powerful AI agents, but the setup can be overwhelming.

I discovered this using Claude Code — it was fantastic, but I was either locked out after 30 minutes on a subscription, or burned 100 kr on the same half hour with API tokens. Frustrating.

With Nexum you skip all of that. We set everything up for you, and you are up and running in two minutes. No technical expertise required.

Cheaper than the big ones

We make it cheaper than the big models by using cheap, simple models for simple tasks — like running commands — and more powerful models where they matter.

The system uses three main models: Organisation (steering the conversation), Utility (executing tasks) and Web browsing (surfing and fetching data). That is how you get Claude-like quality at a fraction of the price.

Privacy

Our hosting is operated from Norway, with Norwegian standards and attitudes to privacy — so model providers do not know who is using the models. We do not pass that data on.

We do not log what you use the models for either, unless we receive mass misuse complaints. Your data is your data. Full stop. As a Norwegian company with servers in Europe, you are covered by GDPR and Norwegian privacy standards.

Freedom

This is an AI computer, not just an AI agent — you can do everything you can do with a PC. It runs Kali Linux by default, and we also have versions optimised for network penetration testing.

You can ask the AI to install any program with Linux (Debian-based) support, as long as it works with text-based commands. The only limit is storage space.

Fair price

This project builds on several open-source forks, so we do not charge for the software itself. We take a small service fee to run servers, set everything up and continuously optimise the service.

We take a modest 20% margin on running costs. Token credits roll over to the next month. We are not here to get rich — we are here to offer a service we love ourselves.

Free choice

We have pre-configured and optimised the agent to use the models that — after extensive testing — work best, especially considering cost. The standard configuration is what we recommend.

But you can also use the very latest Claude models if you want maximum quality. We mark the models we recommend — the choice is yours. It is your agent.

Security

We work continuously to secure the service as much as possible. Every agent is fully isolated from the others, even on the same domain and servers. Think of it as your own VPS.

Isolation is done with Docker containers with separate volumes, separate API keys and separate environment variables. No user can affect another.

Professional features

MCP server, SSH, A2A (agent-to-agent) and much more. The terminal is available in the browser, so you can use the agent's OS directly.

We start from a blank slate with nothing but the standard Kali repositories. Like a new PC: things are slower at first while the agent installs programs and learns to use them — then it just keeps getting faster and more capable.

Projects and memory

Projects can be activated across chats, so you do not have to keep track of which chat is in the right project folder. The agent receives the memories of the active project and treats content differently than when another project is active.

The agent can also call in sub-agents — researchers, developers and more. But we will let you explore that yourself!