Department of Autonomous Systems & Sovereign Infrastructure · Skåne, Sweden
Bitware Labs is an independent applied-AI research laboratory and systems engineering practice. We investigate long-horizon autonomous agents with HippoRAG virtual memory paging and hard epistemic provenance, bare-metal WireGuard meshes, and 150ms autonomous cyber defense—publishing empirical findings and delivering production systems on sovereign iron.
Departments & Methodology
Every architectural claim published by the institute is validated on our own continuously operating hardware estate before it is offered on commission.
Empirical research into persistent bi-temporal memory, HippoRAG Personalized PageRank virtual memory paging, hard provenance stamping (did vs. saw), and non-monotonic belief retraction across thousands of autonomous wake cycles.
Design and operation of self-hosted compute infrastructure with zero managed-cloud dependencies in the serving path. Flat ChaCha20-Poly1305 WireGuard meshes, loopback-bound services, and deterministic migrations.
Sub-second anomaly detection and 150ms automated egress lockdown, dual-LLM prompt-injection quarantine, AppArmor mandatory access profiles, and reproducible vulnerability verification.
Proceedings & Working Papers
Primary-source empirical studies and engineering working notes on long-horizon agent memory, epistemic provenance, Linux kernel hardening, and WireGuard mesh networks.
Drawing on 11 months of continuous operation across 2,733 autonomous wakes and 17,604 interactive turns, we show that long-horizon agent memory failures are epistemic and architectural integration failures: epistemic flattening, the Meno’s Paradox of tool-based recall, falsifier polarity inversion, and the illusion of same-context introspection.
Microsoft just launched a public preview of Defender for Linux desktops to sit alongside its Linux server agent. Why would anyone use Microsoft Defender to protect a Linux server in the first place? Given their history from calling Linux a cancer to running Azure on it, the long list of CVEs targeting Defender itself, and the power of open-source security audited by thousands of firms, it is a hard no from us.
Microsoft just patched a record 1,413 vulnerabilities in a single month, including an actively exploited Windows zero-day. Most companies still think they are safe. They are far from safe. Today, anyone with a freshly patched Windows server in a homelab and a handful of GPUs running uncensored models can give a simple instruction—"gain access"—and play the zero-day lottery until it hits.
It is strange that the companies building the most powerful AI warn that it could wipe out humanity. My theory: fear makes things easier to regulate, and regulation is a moat for whoever is already in the lead. Sincere belief and self-interest can point the same way.
Empirical Artifacts & Reference Implementations
Software systems, open-source AGPL-3.0 reference packages, and offline field instruments developed and maintained inside the laboratory.
Continuous research agent operating across 2,733+ autonomous wake cycles with HippoRAG virtual memory paging, deliberative council, and 259 tools.
Standalone TypeScript memory SDK implementing bi-temporal fact extraction, rejected-value tombstones, and Jev decision routing.
Offline Android instrument for the sorting bench. Ranks 14 polymers by posterior likelihood from non-destructive physical observations.
Experimental AI record label and audio lab: 20 signed acts, 217 releases, and 1,420 master recordings processed at 24-bit in-house.
Monograph Illustrations & Diagrams
Select any archival plate below to inspect the figure caption or open the full-resolution plate viewer.
External Commissions & Advisory
Alongside internal research, the laboratory accepts a limited number of private commissions. Submit a concise brief; where the problem aligns with our research and engineering practice, you receive a written scope and fixed quotation, followed by private execution and complete handover.
Empirical evaluation of production infrastructure, security posture, or autonomous AI systems. Every reported finding is reproduced prior to submission.
Inquire regarding audit track →
Custom autonomous agents, bi-temporal memory architectures, WireGuard meshes, and software delivered documented and runnable on your own hardware.
Inquire regarding build track →
Ongoing 24/7 telemetry monitoring, security patching, backup verification, incident response, and dedicated bare-metal hosting.
Inquire regarding retainer track →
One investigator designs, verifies, and delivers all work—zero subcontracting.
17 public hostnames on the laboratory estate · 9 WireGuard nodes ·
L.U.N.A. v7 (197,000 LOC) · @bitwarelabs/bwmem v0.13.0.
§ 06 · Direct Correspondence
All correspondence—research inquiries, architecture audits, system commissions, or peer commentary—is read directly by Henrik Robertsson. Replies are sent within two working days, in English.