Skip to content

README.md

Latest commit

 

History

History
42 lines (28 loc) · 2.76 KB

README.md

File metadata and controls

42 lines (28 loc) · 2.76 KB

CogOS Research

Part of the CogOS ecosystem -- why it WORKS.

Theoretical foundations, architecture research, and proof-of-concept experiments for cognitive operating system design.

Contents

Document Description
eaefm/thesis.md EA/EFM Thesis -- Externalized Attention and Executive Function Modulation. The core argument: the substrate thinks, the model generates, and quality is a function of boundary quality.
loro/framework.md LoRO Framework -- Low-Rank Observer as a unified abstraction connecting PLE (Per-Layer Embeddings), LoRA (Low-Rank Adaptation), and TRM (Tiny Recursive Model). Three mechanisms, one pattern, operating at different time scales.
poc/ Proof-of-concept experiments (coming soon)
papers/ Academic papers and preprints (coming soon)

What this is

This repo contains the public research that underpins CogOS -- the ideas about why externalizing attention and executive function into a substrate produces better outcomes than scaling model size alone.

The key claims:

  1. EA (Externalized Attention): Deciding what information is relevant before the model sees it -- not retrieval, not augmentation, but selective amplification of what matters.
  2. EFM (Executive Function Modulation): Deciding how the model should behave before it generates -- not prompting, but shaping the computational trajectory through conditioning signals.
  3. LoRO (Low-Rank Observer): PLE, LoRA, and TRM are structurally convergent mechanisms -- all low-rank conditioning of a larger system through a bottleneck. This convergence is not noted in published literature as of April 2026.

What this is not

This is public architecture research related to CogOS. It does not contain the full theoretical framework, fundamental physics, or private workspace internals.

Related projects

  • cogos -- The kernel — continuous process daemon with foveated context and multi-provider routing
  • constellation -- Distributed trust — identity as temporal coherence, O(1) mutual verification
  • mod3 -- Modality bus — translates between thinking and acting, voice-first
  • skills -- Plugin marketplace — Agent Skills across workflow, research, voice, and dev tools
  • charts -- Deployment — Helm charts + Docker Compose
  • desktop -- Native macOS app — kernel management, terminal, dashboard
  • openclaw-plugin -- OpenClaw integration (how it CONNECTS)

License

MIT