Distributed Cognitive Operating System (DCOS)

Version 1.4.2 — A capability-oriented architecture for modular AI orchestration with specialized Cognitive Cells.

DCOS is a research architecture that organizes intelligence as a distributed set of specialized Cognitive Cells instead of one monolithic model. A lightweight Entry Router and a Capability Index select which cells execute a request; cells request capabilities (never other cells) at run time; and a Composer assembles the final response. Cells may use different models, run on different hardware, and evolve independently.

Repository layout

Path	Contents
DCOS_Main_Paper_v1.4.pdf	The research manuscript (Version 1.4.2)
Executive_Summary.pdf	3-page business summary
White_Paper.pdf	Engineering introduction with diagrams
LICENSE.md	License terms (academic vs. commercial use)
NOTICE.md	Dissemination and contact notice
CITATION.cff	Citation metadata
COMPANIES.md	Outreach appendix
repro/	Reproducibility package (installation + benchmarks)
figures/	Architecture and white-paper diagrams
supplementary_data/	Raw benchmark outputs (B1-B6)

Reproducing the benchmarks

The repro/ package contains the prototype source and scripts to reproduce the Chapter 15 benchmarks (B1, B2, B6). It needs only Python 3.9+ and an Ollama server with qwen3:4b-instruct and gemma4:e2b.

cd repro
python src/tests.py --mock          # offline smoke test
python src/benchmarks.py --only b1 b2 b6   # offline benchmark run

See repro/install.md and repro/benchmark.md for details.

Key measured results (prototype, 2026-08)

• B1: DCOS 23.4 s / 1,248 tokens vs. single model 23.9 s / 2,010 tokens; judge 9.0 vs. 7.0.
• B2 (100 real questions): 68% single-cell activation, max depth 4, avg 1.48 cells, 26 capability requests reused, zero errors.
• B5: database_v1 down → auto-served by database_v2, no exception, no edits.
• B6: per-cell Local Memory accumulates across turns; Session Context improves continuity across the conversation.

License

This is not an open-source license. Academic use (research, education, citation) is free with attribution; commercial implementation requires prior written permission. See LICENSE.md and NOTICE.md.

Contact

For research collaboration, technical discussion, or commercial licensing inquiries, please use GitHub Discussions or open an issue in this repository.

Direct contact may also be initiated through the repository owner's GitHub profile.