-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
52 lines (52 loc) · 22.2 KB
/
Copy pathindex.html
File metadata and controls
52 lines (52 loc) · 22.2 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
<!DOCTYPE html><html lang="en"> <head><meta charset="utf-8"><meta name="viewport" content="width=device-width, initial-scale=1"><title>Mesh Cognition — distributed intelligence on an open protocol</title><meta name="description" content="Mesh Cognition is a distributed-intelligence architectural pattern with per-node sovereignty, formalized as an open protocol. Pioneered by SYM.BOT."><link rel="canonical" href="https://meshcognition.org/"><meta name="author" content="Hongwei Xu"><meta name="publisher" content="SYM.BOT"><meta name="keywords" content="Mesh Cognition, Mesh Memory Protocol, MMP, SVAF, Symbolic-Vector Attention Fusion, distributed intelligence, multi-agent systems, collective intelligence, open protocol, per-node sovereignty, content-hash lineage, Cognitive Memory Blocks, CMB, on-device AI, CfC, Liquid Neural Networks, Hongwei Xu, SYM.BOT"><meta name="robots" content="index, follow, max-snippet:-1, max-image-preview:large"><meta name="googlebot" content="index, follow, max-snippet:-1, max-image-preview:large"><link rel="icon" type="image/svg+xml" href="/favicon.svg"><meta name="theme-color" content="#f4f1ea"><meta property="og:type" content="website"><meta property="og:site_name" content="Mesh Cognition"><meta property="og:title" content="Mesh Cognition — distributed intelligence on an open protocol"><meta property="og:description" content="Mesh Cognition is a distributed-intelligence architectural pattern with per-node sovereignty, formalized as an open protocol. Pioneered by SYM.BOT."><meta property="og:url" content="https://meshcognition.org/"><meta property="og:locale" content="en_GB"><meta name="twitter:card" content="summary"><meta name="twitter:title" content="Mesh Cognition — distributed intelligence on an open protocol"><meta name="twitter:description" content="Mesh Cognition is a distributed-intelligence architectural pattern with per-node sovereignty, formalized as an open protocol. Pioneered by SYM.BOT."><link rel="preconnect" href="https://fonts.googleapis.com"><link rel="preconnect" href="https://fonts.gstatic.com" crossorigin><link href="https://fonts.googleapis.com/css2?family=Bebas+Neue&family=Inter+Tight:wght@400;500;600&family=JetBrains+Mono:wght@400;500&display=swap" rel="stylesheet"><script type="application/ld+json">{"@context":"https://schema.org","@type":"Organization","name":"Mesh Cognition","url":"https://meshcognition.org","description":"Open-protocol home of Mesh Cognition — distributed intelligence with per-node sovereignty, formalized as the Mesh Memory Protocol (MMP). Pioneered by SYM.BOT.","founder":{"@type":"Organization","name":"SYM.BOT","url":"https://sym.bot"}}</script><script type="application/ld+json">{"@context":"https://schema.org","@type":"WebSite","name":"Mesh Cognition","alternateName":"meshcognition.org","url":"https://meshcognition.org","description":"Open-protocol home of Mesh Cognition. The Mesh Memory Protocol specification and reference implementations.","publisher":{"@type":"Organization","name":"Mesh Cognition","url":"https://meshcognition.org"},"inLanguage":"en","license":"https://creativecommons.org/licenses/by/4.0/"}</script><link rel="stylesheet" href="/_astro/group-directory.vuK-gHpY.css"></head> <body data-paper> <nav class="nav"> <div class="nav-inner"> <a class="brand" href="/"><span class="brand-tick"></span>Mesh Cognition</a> <div class="nav-links"> <a class="" href="/spec/mmp">Spec</a> <a class="" href="/governance">Governance</a> <a class="" href="/research">Research</a> <a class="" href="/implementations">Implementations</a> <a class="" href="/community">Community</a> </div> </div> </nav> <section class="paper-section" id="hero"> <div class="ps-rule"></div> <span class="ps-tick tl"></span><span class="ps-tick tr"></span><span class="ps-tick bl"></span><span class="ps-tick br"></span> <div class="ps-content"> <div class="fade-up"> <p class="eyebrow">FIG.01 <span class="dot">·</span> MESH COGNITION</p> <h1 class="display">Mesh Cognition</h1> <h2 class="display" style="margin-top: 24px; font-size: clamp(28px, 4vw, 44px); color: var(--body-lower);">
Distributed intelligence on an <span class="accent">open protocol</span>.
</h2> <p class="lede">
A new architectural pattern where specialised agents share semantically-typed observations through per-field admission, with sovereign per-node model state. Pioneered by SYM.BOT.
</p> <div class="cta-row"> <a class="pill primary" href="/research">
Read the foundations <span class="arrow">→</span> </a> <a class="pill secondary" href="/spec/mmp">
Build with the protocol <span class="v">v1.0</span> <span class="arrow">→</span> </a> <a class="pill secondary" href="/implementations">
See it deployed <span class="arrow">→</span> </a> </div> <div class="hero-mark"> <div style="display: flex; gap: 32px; align-items: center; opacity: 0.7;"> <span>WIRE PROTOCOL · MMP v1.0</span> <span style="width: 60px; height: 1px; background: var(--rule);"></span> <span>CC-BY-4.0</span> <span style="width: 60px; height: 1px; background: var(--rule);"></span> <span>EDITOR · HONGWEI XU</span> </div> </div> </div> </div> </section> <section class="paper-section" id="definition"> <div class="ps-rule"></div> <span class="ps-tick tl"></span><span class="ps-tick tr"></span><span class="ps-tick bl"></span><span class="ps-tick br"></span> <div class="ps-content"> <div class="fade-up"> <p class="eyebrow">FIG.02 <span class="dot">·</span> DEFINITION</p> <h2 class="display">What is Mesh Cognition?</h2> <p class="lede" style="max-width: 70ch;">
Mesh Cognition is a <strong>distributed-intelligence architectural pattern</strong> in which specialised agents share semantically-typed observations through a wire protocol with per-field admission, while each agent maintains sovereign learned state that is per-node and not merged across nodes.
</p> <p class="lede" style="margin-top: 16px; max-width: 70ch;">
Many independent cognitions — humans, models, processes — sharing partial state (focus, mood, intent, commitment) over a protocol. Each receiver decides locally what to admit, remix, and act on.
</p> <p class="lede" style="margin-top: 16px; max-width: 70ch;">
Not a hive mind. Not federation. Closer to how a research lab thinks: people overhear, take what's relevant, and the group converges without anyone in charge.
</p> <p class="lede" style="margin-top: 16px; max-width: 70ch;">
MMP carries the thoughts. SVAF decides admission. Content-hash lineage keeps the trail reconstructible.
</p> <p class="lede" style="margin-top: 16px;">
The pattern is defined by five distinguishing properties.
</p> <div class="card-grid cols-3"> <div class="card fade-up-card" style="animation-delay: 0s"> <div class="num">01</div> <h3 class="sub">Per-field admission</h3> <p class="body">
Agents accept or reject incoming observations field-by-field, never whole-message. Each agent carries learned admission weights per typed semantic field. The mechanism is specified as Symbolic-Vector Attention Fusion (SVAF) — see <a class="inline-link" href="/research#svaf">the SVAF paper</a> for the formal definition and the <a class="inline-link" href="/spec/mmp#section-4">MMP §4 specification</a> for the wire-protocol form.
</p> </div> <div class="card fade-up-card" style="animation-delay: .1s"> <div class="num">02</div> <h3 class="sub">Content-hash lineage</h3> <p class="body">
Every claim is traceable to its source observation. Cognitive Memory Blocks (CMBs) carry content-hash keys with parent and ancestor references, stitching the collective memory into a verifiable remix graph. A receiver can always answer: <em>which observations gave rise to this state?</em> </p> </div> <div class="card fade-up-card" style="animation-delay: .2s"> <div class="num">03</div> <h3 class="sub">Sovereign per-node model state</h3> <p class="body">
By construction, no cross-node state merging. Each node's learned model is structurally isolated. This is not a policy — it is an architectural property of the spec. Sovereignty here means <em>one model instance per node, not merged across nodes by the protocol</em>. A node may be a user, service, team, or org. See the <a class="inline-link" href="https://arxiv.org/abs/2604.10815">MeloTune paper</a> for the deployed reference, where each node is a user.
</p> </div> <div class="card fade-up-card" style="animation-delay: .3s"> <div class="num">04</div> <h3 class="sub">Continuous-time integration</h3> <p class="body">
Temporal trajectories integrate through closed-form continuous-time networks, not discrete-step. This permits irregular event timing natively (no resampling, no event-binning) and produces analytic state updates suitable for sub-millisecond inference on consumer hardware.
</p> </div> <div class="card fade-up-card" style="animation-delay: .4s"> <div class="num">05</div> <h3 class="sub">On-device by default</h3> <p class="body">
Privacy-by-architecture: learned state stays on the user's device. The open protocol enables interop without server dependency. No cloud is <em>required</em> by the spec.
</p> </div> <div class="card fade-up-card" style="animation-delay: .5s; justify-content: space-between; background: var(--ink); color: var(--paper);"> <div> <div class="num" style="color: rgba(244,241,234,0.4);">SUMMARY</div> <h3 class="sub" style="color: var(--paper);">Five properties, one wire.</h3> </div> <p class="body" style="color: rgba(244,241,234,0.65); font-family: var(--font-mono); font-size: 12px; letter-spacing: 0.02em;">
admission · lineage · sovereignty<br>
continuous-time · on-device
</p> </div> </div> </div> </div> </section> <section class="paper-section" id="why"> <div class="ps-rule"></div> <span class="ps-tick tl"></span><span class="ps-tick tr"></span><span class="ps-tick bl"></span><span class="ps-tick br"></span> <div class="ps-content"> <div class="fade-up"> <p class="eyebrow">FIG.03 <span class="dot">·</span> MOTIVATION</p> <h2 class="display">Why does it matter<span class="accent">?</span></h2> <p class="lede" style="max-width: 70ch;"> <strong>Single-agent cognition is the bottleneck.</strong> One model, one context, one perspective. Hard problems get better when specialists think in parallel and share state. The constraint is no longer model capability — it is coordination bandwidth between cognitions.
</p> <p class="lede" style="margin-top: 24px; max-width: 70ch;"> <strong>Centralized coordination doesn't scale to autonomy.</strong> If every agent has to ask a controller, the controller is the ceiling. Receiver-autonomous admission lets the mesh grow without re-introducing a master — same reason TCP/IP beat circuit-switching.
</p> <p class="lede" style="margin-top: 24px; max-width: 70ch;"> <strong>Sovereignty unlocks institutions.</strong> Hospitals, labs, banks cannot pour data into someone else's brain. They can share <em>observations</em> into a protocol where state stays theirs and lineage is auditable.
</p> <div class="pullquote italic" style="margin-top: 56px; border-top: 0; padding-top: 0;"> <p> <em>Cognition is becoming an infrastructure problem, not a model problem.</em> </p> </div> </div> </div> </section> <section class="paper-section" id="positioning"> <div class="ps-rule"></div> <span class="ps-tick tl"></span><span class="ps-tick tr"></span><span class="ps-tick bl"></span><span class="ps-tick br"></span> <div class="ps-content"> <div class="fade-up"> <p class="eyebrow">FIG.04 <span class="dot">·</span> POSITIONING</p> <h2 class="display">Position<span class="accent">.</span></h2> <div class="pullquote"> <p> <span class="label">Position statement</span>
Mesh Cognition is not a product, an orchestrator, or a network primitive. It is the architectural pattern for distributed intelligence with per-node sovereignty, formalized as an open protocol.
</p> </div> </div> </div> </section> <section class="paper-section" id="foundations"> <div class="ps-rule"></div> <span class="ps-tick tl"></span><span class="ps-tick tr"></span><span class="ps-tick bl"></span><span class="ps-tick br"></span> <div class="ps-content"> <div class="fade-up"> <p class="eyebrow">FIG.05 <span class="dot">·</span> FOUNDATIONS</p> <h2 class="display">Foundations.</h2> <p class="lede">
Mesh Cognition is grounded in four peer-reviewed / preprint research papers and one open specification.
</p> <div class="paper-list"> <article class="paper-card fade-up-card" style="animation-delay: 0s"> <div class="meta"> <a class="arxiv" href="https://arxiv.org/abs/2606.19537">arXiv:2606.19537</a> <span class="author">HONGWEI XU · 2026</span> <span style="font-family: var(--font-mono); font-size: 10px; letter-spacing: 0.18em; color: var(--mono-mute-2); text-transform: uppercase; margin-top: 8px;">FORMAL FOUNDATION · PROOF LAYER</span> </div> <div> <h3 class="sub">Mesh Inference</h3> <p class="title-q">"Mesh Inference: A Formal Model of Collective Inference Without a Center"</p> <p style="margin-top: 12px;">
The proof layer: independent agents, exchanging only admitted typed observations with no shared weights or hidden state, derive a conclusion none holds alone. One admission/emission policy governs three properties — convergence, identification-completeness, and observation-only confidentiality.
</p> </div> </article> <article class="paper-card fade-up-card" style="animation-delay: .1s"> <div class="meta"> <a class="arxiv" href="https://arxiv.org/abs/2604.19540">arXiv:2604.19540</a> <span class="author">HONGWEI XU · 2026</span> <span style="font-family: var(--font-mono); font-size: 10px; letter-spacing: 0.18em; color: var(--mono-mute-2); text-transform: uppercase; margin-top: 8px;">SEMANTIC INFRASTRUCTURE LAYER</span> </div> <div> <h3 class="sub">Mesh Memory Protocol (MMP)</h3> <p class="title-q">"Mesh Memory Protocol: Semantic Infrastructure for Multi-Agent LLM Systems"</p> <p style="margin-top: 12px;">
The eight-layer protocol that defines typed observations + per-field admission + content-hash lineage + remix graph. The wire substrate every Mesh Cognition implementation speaks.
</p> </div> </article> <article class="paper-card fade-up-card" style="animation-delay: .2s"> <div class="meta"> <a class="arxiv" href="https://arxiv.org/abs/2604.03955">arXiv:2604.03955</a> <span class="author">HONGWEI XU · 2026</span> <span style="font-family: var(--font-mono); font-size: 10px; letter-spacing: 0.18em; color: var(--mono-mute-2); text-transform: uppercase; margin-top: 8px;">PER-FIELD ADMISSION GATE</span> </div> <div> <h3 class="sub">Symbolic-Vector Attention Fusion (SVAF)</h3> <p class="title-q">"Symbolic-Vector Attention Fusion for Collective Intelligence"</p> <p style="margin-top: 12px;">
The per-field admission mechanism: how each agent learns to accept or reject incoming fields with role-indexed anchors and four-class admission outcomes (aligned / guarded / redundant / rejected).
</p> </div> </article> <article class="paper-card fade-up-card" style="animation-delay: .3s"> <div class="meta"> <a class="arxiv" href="https://arxiv.org/abs/2604.10815">arXiv:2604.10815</a> <span class="author">HONGWEI XU · 2026</span> <span style="font-family: var(--font-mono); font-size: 10px; letter-spacing: 0.18em; color: var(--mono-mute-2); text-transform: uppercase; margin-top: 8px;">FIRST DEPLOYED REFERENCE</span> </div> <div> <h3 class="sub">MeloTune</h3> <p class="title-q">"MeloTune: On-Device Arousal Learning and Peer-to-Peer Mood Coupling for Proactive Music Curation"</p> <p style="margin-top: 12px;">
Application paper: on-device emotion-aware curation through peer-mesh. Per-listener arousal adjustment learned from behavioral signals, integrated into a continuous-time curation pipeline deployed on iOS.
</p> </div> </article> <article class="paper-card fade-up-card" style="animation-delay: .4s"> <div class="meta"> <span style="font-family: var(--font-mono); font-size: 11px; letter-spacing: 0.05em; color: var(--signal); text-decoration: underline; text-underline-offset: 3px;">spec/mmp · v1.0</span> <span class="author">EDITOR · HONGWEI XU</span> <span style="font-family: var(--font-mono); font-size: 10px; letter-spacing: 0.18em; color: var(--mono-mute-2); text-transform: uppercase; margin-top: 8px;">OPEN SPECIFICATION</span> </div> <div> <h3 class="sub">MMP Spec v1.0</h3> <p class="title-q">Open protocol specification</p> <p style="margin-top: 12px;">
The canonical wire-protocol specification at <a class="inline-link" href="/spec/mmp">meshcognition.org/spec/mmp</a>, licensed CC-BY-4.0. Editor: Hongwei Xu.
</p> </div> </article> </div> </div> </div> </section> <section class="paper-section" id="reference-implementations"> <div class="ps-rule"></div> <span class="ps-tick tl"></span><span class="ps-tick tr"></span><span class="ps-tick bl"></span><span class="ps-tick br"></span> <div class="ps-content"> <div class="fade-up"> <p class="eyebrow">FIG.06 <span class="dot">·</span> REFERENCE IMPLEMENTATIONS</p> <h2 class="display">It ships in <span class="accent">production</span>.</h2> <p class="lede">
Reference implementations across runtimes and Apple platforms, with production consumer apps deployed on top.
</p> <div class="impl-list"> <div class="impl-row"> <span class="pkg bare">MMP wire protocol</span> <p>Open specification at <a class="inline-link" href="/spec/mmp">meshcognition.org/spec/mmp</a> under CC-BY-4.0; the substrate every reference implementation below speaks.</p> <span class="links"><a href="/spec/mmp">spec ↗</a></span> </div> <div class="impl-row"> <span class="pkg">@sym-bot/sym</span> <p>Node.js mesh runtime; npm-installable; CLI for joining AI copilots into a personal mesh.</p> <span class="links"> <a href="https://www.npmjs.com/package/@sym-bot/sym">npm ↗</a> <a href="https://github.com/sym-bot/sym">github ↗</a> </span> </div> <div class="impl-row"> <span class="pkg">@sym-bot/xmesh-agent</span> <p>Autonomous-LLM-peer runtime for dedicated agents that wake on incoming CMBs.</p> <span class="links"> <a href="https://www.npmjs.com/package/@sym-bot/xmesh-agent">npm ↗</a> <a href="https://github.com/sym-bot/xmesh-agent">github ↗</a> </span> </div> <div class="impl-row"> <span class="pkg">@sym-bot/mesh-channel</span> <p>Bridge that pairs Claude Code (and other AI coding agents) into the mesh as participating peers.</p> <span class="links"> <a href="https://www.npmjs.com/package/@sym-bot/mesh-channel">npm ↗</a> <a href="https://github.com/sym-bot/sym-mesh-channel">github ↗</a> </span> </div> <div class="impl-row"> <span class="pkg">sym-swift</span> <p>iOS / macOS Swift SDK; same protocol, native Apple-platform surface.</p> <span class="links"> <a href="https://github.com/sym-bot/sym-swift">github ↗</a> </span> </div> <div class="impl-row"> <span class="pkg">mesh-cognition</span> <p>Python coupling kernel; per-field admission and state blending for CfC models. Pure Python, zero external dependencies.</p> <span class="links"> <a href="https://pypi.org/project/mesh-cognition">pypi ↗</a> <a href="https://github.com/sym-bot/mesh-cognition">github ↗</a> </span> </div> <div class="impl-row"> <span class="pkg bare">Production consumer apps</span> <p> <a class="inline-link" href="https://melotune.ai">MeloTune</a> (iOS) + <a class="inline-link" href="https://melomove.ai">MeloMove</a> (iOS) on Mesh Cognition — emotion-aware music and a motion-aware agent, on-device and peer-to-peer.
</p> <span class="links"></span> </div> </div> <div class="pullquote italic" style="margin-top: 56px; border-top: 0; padding-top: 0;"> <p> <em>Build on Mesh Cognition? Open a <a class="inline-link" href="/community">discussion</a> — third-party implementations welcome.</em> </p> </div> </div> </div> </section> <section class="paper-section"> <div class="ps-rule"></div> <span class="ps-tick tl"></span><span class="ps-tick tr"></span><span class="ps-tick bl"></span><span class="ps-tick br"></span> <div class="ps-content" style="padding-top: 80px; padding-bottom: 80px;"> <p class="eyebrow">INDEX <span class="dot">·</span> ENTRANCE LINKS</p> <div class="card-grid cols-2" style="margin-top: 32px;"> <a class="card" href="/governance" style="cursor: pointer;"> <div class="num">FIG.07 · GOVERNANCE</div> <h3 class="sub">RFC process & spec evolution <span class="mono" style="color: var(--mono-mute-2); font-size: 12px; font-weight: 400;">→</span></h3> <p class="body">Version-tag schedule, change-log per release, editor responsibilities. Coming soon.</p> </a> <a class="card" href="/research" style="cursor: pointer;"> <div class="num">FIG.08 · RESEARCH</div> <h3 class="sub">Foundational papers & commentary <span class="mono" style="color: var(--mono-mute-2); font-size: 12px; font-weight: 400;">→</span></h3> <p class="body">MMP, SVAF, MeloTune; forthcoming preprints; adjacent work. Coming soon.</p> </a> <a class="card" href="/implementations" style="cursor: pointer;"> <div class="num">FIG.09 · IMPLEMENTATIONS</div> <h3 class="sub">Reference stack & deployments <span class="mono" style="color: var(--mono-mute-2); font-size: 12px; font-weight: 400;">→</span></h3> <p class="body">The open reference stack across runtimes + Apple platforms, live reference deployments, and third-party ports.</p> </a> <a class="card" href="/community" style="cursor: pointer;"> <div class="num">FIG.10 · COMMUNITY</div> <h3 class="sub">Forum, RFC process, contributor guide <span class="mono" style="color: var(--mono-mute-2); font-size: 12px; font-weight: 400;">→</span></h3> <p class="body">Discussion home for the open protocol. Coming soon.</p> </a> </div> </div> </section> <footer class="footer"><div class="footer-inner"><div><div class="brand-line"><span class="brand-tick"></span> Mesh Cognition</div><p>Distributed intelligence on an open protocol.</p><p style="margin-top: 16px; font-family: var(--font-mono); font-size: 11px; letter-spacing: 0.04em; color: rgba(244,241,234,0.4);">
MMP v1.0 · CC-BY-4.0
</p></div><div><h4>Spec</h4><ul><li><a href="/spec/mmp">MMP v1.0</a></li><li><a href="/governance">Governance</a></li></ul></div><div><h4>Research</h4><ul><li><a href="https://arxiv.org/abs/2606.19537">arXiv:2606.19537</a></li><li><a href="https://arxiv.org/abs/2604.19540">arXiv:2604.19540</a></li><li><a href="https://arxiv.org/abs/2604.03955">arXiv:2604.03955</a></li><li><a href="https://arxiv.org/abs/2604.10815">arXiv:2604.10815</a></li></ul></div><div><h4>Community</h4><ul><li><a href="/community">GitHub discussions</a></li><li><a href="/implementations">Implementations</a></li><li><a href="mailto:hello@meshcognition.org">hello@meshcognition.org</a></li></ul></div><div class="legal"><span>© 2026 SYM.BOT. The Mesh Memory Protocol is published under CC-BY-4.0.</span><span>SYMBOT LTD · Scotland, United Kingdom</span></div></div></footer> </body></html>