Nare Labs

MFLM: A sub-quadratic alternative to Transformers using causal mass-field convolutions for infinite context and O(N) complexity.

An autonomous AI coding agent that lives in your terminal.

Installer for Sharrowkin IDE

DANIUS-1: Teaching frozen language models to think in shapes and grids, not words. A project by Nare Labs.

A decoupled context-compression co-processor utilizing gated cross-attention for linear-time, O(1)-memory sequence scaling in frozen large language models. (159 символов)

Sparse attention for frozen LLMs. Train a tiny router that picks the top-K keys per query, swap dense attention for sparse — no retraining the model.

Static-Q Transformer: replace self-attention's content-based query projection with a learnable per-position table. Empirical study of where this works and where it breaks across model sizes.

Capability-Augmented Generation through Reusable Graph — a research framework for accumulating and composing executable LLM skills.

s-git: Semantic version control that tracks code meaning and AST structure, not text.