neurons.me

.me is a deterministic, semantic, reactive kernel that runs locally in the browser. No cloud. No registration. No vendor lock-in. The problem is Contextual Momentum: coherence is not correctness. A representation can begin perfectly correct — an accurate observation, faithfully recorded — and still go wrong, because the world moves and the representation does not. A model answering from memory says the total is 65, because it remembers bread at 20. The text stayed coherent with itself. It did not stay coherent with the world. .me fixes this by separating proposing from deciding. Muse proposes. .me decides. Muse proposes facts, rules, and changes in .me syntax. The kernel validates them, dry-runs them, and writes them only after the user confirms. If a path is unknown, the proposal is rejected at validation. Nothing is written. Nothing to confirm. The graph is unchanged. Hallucination never becomes state. State is a graph, not text. Facts are stored. Derivations are recomputed: me.menu.coffee.price(45) me.menu.bread.price(20) me.order["="]("total", "menu.coffee.price + menu.bread.price") me.menu.bread.price(25) // total derives to 70 No recall. No guessing. With 16% tax and a split between Ana and Luis, the total derives to 81.2, with full provenance via me.explain(). It scales because it is O(k): cost scales with what changed, not with graph size. In the demo, 300 orders and 903 paths: one write (orders[7].bread 20 → 25) recomputes only k = 3; a tax.rate change (1.16 → 1.10) recomputes k = 302. Published benchmark, this.me 4.2.0: one write in a 5,000-node graph, k = 2, 0.17 ms. Best of all, it runs locally. Recompute burns zero tokens. .me works with any LLM; shown here with muse-spark-1.3. Open source: github.com/neurons-me/.me Live demo: https://neurons-me.github.io/.me/LLMS/Meta/ Note: the this.me kernel and .GUI existed before. Built this week: LLMS chat, Muse adapter, validator.