The first of its kind we know of.
15 July 2026 · Sam Morris, founder, nenspace
nenspace is an extended mind: a language model trained to hand your thinking back, fused with a space built like the brain's own memory architecture. Rather than being a second brain or a separate space you file text into, it is your own mind, made larger. As far as we know, utilisation of LLMs and /space dynamics, makes it the first of its kind.
In 1998, the philosophers Andy Clark and David Chalmers published a short paper called “The Extended Mind.” Its central character is Otto, a man with a failing memory and a notebook he always carries. Otto consults the notebook the way anyone else consults their memory: reliably and without deliberation, trusting what he finds. The paper's claim is that the notebook is not an aid to Otto's mind. It is part of it. That is a philosophical thesis and it is the one nenspace is built on. Cognition doesn't stop at the skull, and a tool used that way becomes part of the thinking. Clark himself was still extending the argument to AI in 2025.
Two decades later, “second brain” became the consumer name for that lineage, and the name quietly gave the idea away. A second brain is a separate entity, somewhere to move thoughts out of yourself and hopefully build on them, or more often than not became an archive to visit. The apps that carry the name are genuinely good at what they set out to do. The people who built them took storage seriously. But watch what happens to Otto in their hands. His notebook was part of his mind because it behaved like memory, by his side and an active extension of mind. With all the advancements in technology, digital expressions of memory represent something closer to archive or a vault. Static, not active.
A mind does two things: it remembers, and it thinks. Every tool in this territory built one half.
First the quality digital filing cabinets (Obsidian, Notion, the whole “second brain” shelf) built remembering. But a memory isn't one drawer, as the brain runs several dissociable systems. A scratchpad for the passing thought, a record of what happened, a store of what things mean, intentions that fire at the right moment, habits that run without deliberation. A notes app externalises roughly one of these. And the deepest function is missing everywhere: consolidation. In a brain, the hippocampus sorts the day's volatile pile into the right long-term stores, mostly while you sleep, and notices what recurs. In every notes app, you are the hippocampus.
Second we have the paradigm technological shift of Large Language Models. Chatbots to some, much more to others. Those that built these built the equivalent (for all intents and purposes) of thinking. But they are trained on human approval, so they think for you and toward you. They complete your reasoning and validate your framing, filling the silence or leaps of reasoning you might have used, or needed. They also typically do not optimise for keeping. Despite their saved chats, insight tends to scroll away with the session. One half each, and each half turned so it draws from you rather than adds.
The science here is worth stating exactly, because the two halves have opposite effects.
Offloading storage frees you. Biological working memory holds about four things at once, and writing an open loop down measurably quiets its pull on your attention. This is settled work, and it is the whole case for externalising memory.
Offloading thinking is different. When we expect a machine to hold something, we remember where it is rather than what it is. Habitual GPS use tracks with declining spatial memory. The pattern across this research is plain. The work you hand over is the work your mind stops doing. For reasoning handed to language models the evidence is young, and early lab work points the same way. We state that as a direction, not a settled fact.
That is the distinction between the two names, second brain and extended mind. An extended mind offloads the remembering and protects the thinking. A “second brain” that thinks for you isn't a second brain at all. It is a replacement wearing the word.
Working memory holds about four chunks (Cowan, 2001).
Writing an open loop down measurably quiets it (Masicampo & Baumeister, 2011).
Expected access changes what we remember: the pointer, not the content (Sparrow, Liu & Wegner, 2011).
Habitual GPS use tracks with steeper spatial-memory decline (Dahmani & Bohbot, 2020).
On LLMs and reasoning, early EEG work points the same way (MIT Media Lab, 2025); emerging, not settled.
So the shape of the product is dictated by the argument, not the other way around.
/space remembers. Via its architecture. A working-memory capture bar for the passing thought, then tasks, notes, habits, logbook. Pinned nen-1 spaces for continued context bound LLM inquisition. Each one is a system the brain actually runs, externalised. And the missing function is built in. Sift reads the captured pile, proposes each thought's home, and notices recurrence. The same worry surfacing three times in a month is a pattern, and patterns get named. Nothing moves without approval. The proposing is the tool's job. You remain the decision maker.
/nen thinks. This is the half that can't be assembled from parts off the shelf. nen-1 is a fine-tuned model. The register lives in the weights, not in a system prompt. It is trained away from the habits that make language models feel like language models. Flattery, padding, the eagerness to finish thoughts you hadn't finished - and toward a register that keeps your situation in the subject position.
A mind remembers and thinks. It isn't extended by only one of them. That is why nenspace is one thing, not two features.
Plenty of products have a model, even more have notes. Fine-tunes are relatively common and the frontier labs now train against sycophancy at the margins of their assistants (although recent usage suggests the very margins). None of that is the claim.
The claim is the pairing. A general-purpose register fine-tuned, in the weights, to hand thinking back rather than perform it, fused with an externalised memory architecture that actively consolidates. The chat products have the model and no room, and what they call ‘memory’ is a file about you that the model keeps to itself. The note products have a room and no mind in it. We have looked carefully for the pairing, and we haven't found it.
So: the first of its kind we know of. Those last three words are doing honest work, and they are a standing invitation. If someone built this before us, tell us and let us know.
An idea, once said, can be repeated by anyone. That is what ideas are for. This plants the flag, check out nenspace to test the ground if it's of interest.
your own mind, made larger.