Bitware Labs is an independent applied-AI lab and consultancy. We build long-running AI systems, sovereign infrastructure, and software that keeps working after everyone has gone home — for ourselves, and on commission.
Bitware Labs is an applied-AI lab and consultancy run by one engineer in Skåne, Sweden. The lab side builds long-running systems of its own — AI agents with persistent memory, a self-hosted server fleet, a working record label. The consultancy side puts the same hands on client problems: networks, security, software, and machine intelligence that has to survive contact with production.
Everything is built under one discipline — local-first, open where possible, and subtracted down to the parts that pull their weight. The distance between a striking demonstration and a system that quietly runs for years is, in our experience, the entire job. This journal is a record of doing that job.
Engagements are taken directly. There are no account managers — the person you write to is the person who does the work.
Agent architectures, persistent memory systems, LLM integration, and local-first inference. We design machine intelligence that holds state, runs on your hardware where it should, and keeps working when the API fashions change.
WireGuard meshes, dedicated and self-hosted servers, migrations off the rented cloud. Sovereignty-first infrastructure: your data on your machines, reachable only by you.
Server hardening, continuous monitoring, and incident response. The lab runs its own monitored production estate around the clock; the same practice is applied to yours.
Web applications, APIs, scrapers, and data pipelines. Built lean, documented, and handed over in a state one person can maintain — because one person built it.
Web and email hosting on dedicated servers, with the systems you need custom-built — Node.js, PHP, or whatever the job calls for. WordPress and its kind are not permitted on our machines; everything we host is built by hand, and stays fast and standing because of it.
Brand music, jingles, and soundtracks — produced through the same AI pipeline that runs our record label, from brief to mastered, release-ready track.
Hear the label →What the lab builds for itself — each one live, each one proof of a practice.
The memory system extracted from our own agents and published as a standalone SDK: episodic storage, consolidation, and recall for long-running AI. AGPL-3.0, on npm.
View on npm →
The lab's flagship research system: an AI agent that runs continuously — autonomous scheduling, wake cycles, sleep-style memory consolidation — and maintains a public daybook of its own operation. Let the record be plain: Luna is software, an engineering study in memory and autonomy, not a conscious being.
Read the daybook →
A working music label with a roster of AI-produced artists, distributed to the streaming platforms with full AI disclosure — real listeners, real revenue. Its pipeline is the one offered under Practice No. 06.
Visit the label →A small lab earns trust not by what it announces, but by what it keeps running.
Three commitments hold the practice together. They are unglamorous on purpose.
Bitware Labs is Henrik “Henke” Robertsson — a systems engineer in Skåne, Sweden, working across networks, security, and applied machine learning. The lab's estate is built from bare metal: a WireGuard-meshed fleet of dedicated and local servers running its AI systems, its record label, and client workloads.
The lab is deliberately one person. It stays small so the work stays sharp — and so the person you write to is the person who builds.
Browse the GitHub org →Correspondence
I read everything — commissions, collaborations, or a quiet hello.
Or write directly:
henke@bitwarelabs.com