Reference · v2.0.0 · Reference Implementation

Documentation

A complete reference for AetherMesh — architecture, formal definitions, CRDTs, gossip, and BFT

Published research with reference implementation. AetherMesh is at v2.0.0. The manuscript is published on Zenodo. A reference implementation is built (107 tests, 79% coverage) and 3 experiments are completed.

Table of Contents

Overview

AetherMesh is a fully decentralized peer-to-peer synchronization framework for autonomous agent networks that modify shared codebases and metadata DAGs. It replaces locking and central coordination with a tripartite architecture, and is the decentralization layer (fourth layer) of the Aether family, above AetherGit (storage), AetherFlow (integration), and AetherProof (verification).

FieldValue
Full titleAetherMesh: Decentralized Peer-to-Peer Agent Synchronization and State CRDTs for Autonomous Multi-Agent Systems
Version2.0.0 (Reference Implementation) / 1.0.0 (Formal Architecture)
StageReference Implementation · Reproducible Experiments
LicenseMIT (code) · CC-BY-4.0 (docs, manuscript)
DomainDistributed Systems / Software Engineering / Formal Methods
DOI (v2.0.0)10.5281/zenodo.24053492
Concept DOI10.5281/zenodo.24038037
OSF10.17605/OSF.IO/9PE5V
PyPIaethermesh
Record IDAETHERMESH-02

"The simulation results here are an invitation to such evaluation, not a substitute for it."

Formal Definitions

Definition 1 — Convergent State Transition. Let A = {a1, a2, …, an} be a set of autonomous agents. Each agent maintains a local replica state Ra(t) at logical time t. A transition caused by local execution or gossip reception is defined as:

R_a(t+1) = R_a(t) ⊔ ⨆_{m ∈ M_a} m

Where Ma is the set of incoming mutation payloads that passed admission during (t, t+1). By commutativity, associativity, and idempotence, the order and multiplicity of payloads do not matter.

Definition 2 — CRDT State Space. The state space is a bounded join-semilattice:

(S, ⊔, ⊥)  bounded join-semilattice,  s_reconciled = s_i ⊔ s_j

where ⊔ is binary, commutative, associative, and idempotent.

Definition 3 — Gossip Infection Probability. The probability that a node remains uninfected after t gossip steps:

P_uninfected(t) ≈ e^(−β·t)

β is the transmission rate constant determined by fanout k and exchange frequency r.

Definition 4 — Byzantine Admission Rule. Before a peer accepts a remote CRDT delta Δi, local verification daemons evaluate cryptographic provenance, structural invariants, and semantic compatibility matrices. Only deltas bearing valid signatures and satisfying local predicates are admitted.

Architectural Pillars

PillarDescriptionStatus
1 — CRDTs for Code and Metadata DAGsState is the set of admitted deltas; set union is the join. Delta-state replication by default; state-based fallback. Metadata uses product lattice (OR-sets, LWW registers, grow-only maps).Implemented
2 — Epidemic Gossip ProtocolsFive-step anti-entropy exchange: digest → digest + difference list → delta request → delta batch → local verification and join. Uniform random peer selection over a partial view.Implemented
3 — Lightweight Byzantine Fault ToleranceValidity filtering, deliberately not a consensus protocol: (1) signature, (2) DAG integrity, (3) schema, (4) semantic compatibility matrix. Missing parents are deferred, not rejected.Implemented

Admission Pipeline (Figure 7). A remote delta passes through four checks before joining the local semilattice:

Admission determinism requirement (A6). Each admission predicate must be a pure, deterministic function of the delta and of its causal closure — not of arrival order, local clock, or non-causal replica contents. Local checks such as per-origin rate limits are treated as delivery policies that may delay a delta but never permanently reject it.

Experiments (v2.0.0)

#ExperimentResult
1Convergence LatencyConverged at 1–2 rounds (10–500 nodes)
2Conflict Resolution0 deadlocks, 0 conflicts flagged
3Fault Resilience100% block rate (10–500 malformed)
Experiments use simulated data. No real-world deployment has been evaluated.

Validation

CheckResult
Formal Definitions4 (all defined)
Theorems3 (stated with proof sketches)
Assumptions6 (A1–A6)
Architectural Pillars3 (all specified)
Figures5 (all generated)
References15
Open Questions50 (Q1–Q50)
Design Decisions52 (D1–D52)
Threats15 (T1–T15)
ImplementationBuilt (src/aethermesh/, 15 modules)
Tests107 passed, 79% coverage
Experiments3/3 completed (simulated data)
Real-world deploymentNot evaluated
Causal ClaimsNone (formal analysis only)

Scope & Limits

What AetherMesh supports:

What AetherMesh does not support:

Design philosophy. AetherMesh is a research paper with a reference implementation, not a production system. Its purpose is to specify an architecture precise enough to be tested against actual agentic workloads — and to make the limits of that specification explicit.

Citation

APA: Baladi, S. (2026). AetherMesh: Decentralized Peer-to-Peer Agent Synchronization and State CRDTs for Autonomous Multi-Agent Systems (Version 2.0.0). Zenodo. https://doi.org/10.5281/zenodo.24053492

@techreport{baladi2026aethermesh,
  author       = {Baladi, Samir},
  title        = {AetherMesh: Decentralized Peer-to-Peer Agent
                  Synchronization and State CRDTs for Autonomous
                  Multi-Agent Systems},
  institution  = {Ronin Institute / Rite of Renaissance},
  type         = {Research Paper},
  year         = {2026},
  month        = {November},
  version      = {2.0.0},
  doi          = {10.5281/zenodo.24053492},
  url          = {https://doi.org/10.5281/zenodo.24053492}
}

Software (Reference Implementation):

@software{baladi2026aethermesh_v2,
  author       = {Baladi, Samir},
  title        = {{AetherMesh} v2.0.0: Reference Implementation of
                  Decentralized Peer-to-Peer Agent Synchronization
                  and State CRDTs},
  year         = {2026},
  month        = {November},
  publisher    = {PyPI},
  version      = {2.0.0},
  doi          = {10.5281/zenodo.24053492},
  url          = {https://pypi.org/project/aethermesh/}
}
Status note: A reference implementation is built (src/aethermesh/, 107 tests, 79% coverage) and 3 experiments are completed. Experiments use simulated data; no real-world deployment is evaluated. Theorems T1–T3 remain proof sketches, not machine-checked.