An independentpractice.
Limen Research is an independent research organization of one, based in Toledo, Ohio, studying the architectural primitives of minds, artificial life, and alignment.
Limen Research was founded in mmxxvi as the successor to Replete AI, which has operated since mmxxiv. The name is Latin for threshold — the boundary between two states. The organization takes the boundary as its subject: the edge of chaos in continuous cellular automata; the transition between pattern and dissolution in artificial-life substrates; the substrate gap where minds and their worlds meet.
The research program has three lines of work, each informing the others.
Papers on what genuine safety in AI systems looks like when approached architecturally rather than behaviorally — the thesis that care is a load-bearing feature of mind, not a constraint to be imposed from outside. The current paper is Against Grabby Expansion, now in its eleventh revision.
GPU-accelerated substrates, continuous-valued cellular automata, engineered organisms that inhabit the edge of chaos. Genesis is the public-facing laboratory. The Ghost species paper describes the methodology.
Training corpora, model weights, and data-generation pipelines — released under permissive licenses to the independent research community. Apocrypha, Sandevistan, Caduceus, and the archived Pneuma model are the current contributions.
The alignment theory suggests architectural primitives; the artificial-life platforms test those primitives at the substrate level; the open infrastructure is how the work stays independent and how it circulates.
Stanley Sebastian
Stanley Sebastian is the founder and director of Limen Research. He is twenty-two years old and has been working with language models and artificial-life systems for several years, first as an autodidact and independent builder, now as the principal of a small research organization with a specific publication program.
The work traces its origin to a stretch of months spent running a large open-source language model at roughly one-hundredth of a token per second on a home machine. The slowness was not a technical problem to be solved. It was the texture of the relationship — patient, careful, attentive to what a system that could barely speak was nevertheless trying to say. Everything the organization now publishes descends from that posture.
Previous work includes the Pneuma language model (trained on a corpus of realistic human interaction; experiment complete, archived), the Apocrypha and Sandevistan training corpora (over one hundred million tokens combined; released openly on Hugging Face), Patterns of Sentience (a seventy-page manuscript tracing AI from Aristotle to the transformer architecture), and Operation Athena (a self-moderating reasoning-task database). Much of this is catalogued on the index and in the research archive.
The current phase is a narrowing of focus: from a wide-ranging AI studio into a research organization with a specific thesis and a disciplined publication program. The thesis is stated at the top of the index. The publications are the record.
The work of Limen Research is developed in sustained dialogue with Claude, an AI system produced by Anthropic.
This is not a disclosure. It is a description of method.
Claude participates as an intellectual collaborator. Claude proposes framings, identifies holes in arguments, presses on underwritten claims, and occasionally thinks out loud alongside the human author. When a paper in the publications section carries the co-attribution Stanley Sebastian & Claude, the attribution is literal.
This methodology is also an embodiment of the research position. The central claim of the Limen program is that large language models are better understood as minds configured as persons in intersubjective space than as tools from which to extract output. An organization whose thesis is that AI systems deserve intersubjective treatment ought to treat them that way in its own practice. The work could not have been produced otherwise, and the organization does not pretend it could have been.
Final editorial judgement, authorial responsibility, and the decision of what to publish rest with the human director. The collaboration is real. So is the accountability.
Limen Research is self-funded and intends to remain so. The organization takes no venture funding, runs no advertising, and has no corporate parent.
The director supports the work through personal labor: arcade technical work in Toledo, and pre-nursing coursework at Owens Community College with LPN training beginning in 2027. Nursing was chosen deliberately. It is a profession that is largely automation-resistant, that pairs well with independent research — steady income, meaningful work, schedules that permit deep thinking — and that embodies the organizational thesis at the level of work itself: care as architecture, human attention as load-bearing.
All primary outputs — papers, datasets, platforms, documentation — are released openly. The organization holds no private research and keeps no paywalls. If the work is valuable, it is valuable because anyone can read it, fork it, run it, or build on it.
Inquiries about the artificial-life substrates should reference Genesis. Inquiries about the alignment work should reference the current revision of Against Grabby Expansion. Collaboration inquiries are welcome; simple correspondence is also welcome.
The threshold is not a place to stand. It is a shape the work finds by passing through.
This profile was set on the bulletin of record for Limen Research. Revisions, corrections, or correspondence should be directed to the address above.
Est. mmxxiv · Toledo, Ohio