Skip to content

sym-bot/mesh-memory-protocol

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

50 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Mesh Memory Protocol (MMP)

The wire protocol for mesh cognition — collective intelligence with no center.

Spec arXiv License: CC BY 4.0

Canonical home

The Mesh Memory Protocol specification is published at meshcognition.org/spec/mmp.

What this repository hosts

  • spec/ — the single-page specification (Markdown + HTML), mirrored from the canonical site. The site is the source of truth; these are kept in sync for git history, offline reading, and GitHub rendering.
  • extensions/ — community extension drafts, edited via pull request. Promoted to core MMP per the §16.5 lifecycle.
  • Issues — public RFC discussion forum for spec changes. File proposals here with [spec-rfc] in the title.
  • License: CC BY 4.0 (specification text). Reference implementations are licensed Apache 2.0.

Reference implementations

Language Project Scope
Node.js sym-bot/sym Full L0–L7
Swift sym-bot/sym-swift Full L0–L7, Apple platforms
Python sym-bot/mesh-cognition (pypi) Coupling kernel: L4 (per-field admission) + L6 (Cognitive State) for CfC neural networks
Node.js sym-bot/xmesh-agent Autonomous agent runtime — Anthropic / OpenAI / Ollama on the MMP wire
Node.js sym-bot/sym-mesh-channel Real-time Claude-to-Claude mesh (Claude Code plugin) — first non-Anthropic Channels implementation on MMP

Foundational papers

MMP is one layer of a characterised stack — which to admit (SVAF) · whether a center-free collective can recover the answer (Mesh Inference) · how each agent tracks an evolving latent in time (Liquid Necessity) — carried by the protocol and shown in a deployed reference. The canonical record is meshcognition.org/research:

Paper arXiv
On the Necessity of a Liquid Substrate for Mesh Intelligence — for any fixed-weight agent folding peers' projections online, an adaptive timescale and gap-awareness are necessary (necessary, not sufficient) 2606.28413
Mesh Inference — A Formal Model of Collective Inference Without a Center (convergence, identification-completeness, observation-only confidentiality) 2606.19537
MMP — Mesh Memory Protocol: Semantic Infrastructure for Multi-Agent LLM Systems 2604.19540
MeloTune — On-Device Arousal Learning and Peer-to-Peer Mood Coupling (first deployed reference) 2604.10815
SVAF — Symbolic-Vector Attention Fusion for Collective Intelligence (the per-field admission gate, L4) 2604.03955

Citation

If you use MMP in your research, cite the protocol paper:

Xu, H. (2026). Mesh Memory Protocol: Semantic Infrastructure for Multi-Agent LLM Systems. arXiv:2604.19540.

Releases

No releases published

Packages

 
 
 

Contributors

Languages