Distributed Systems · Formal Methods · Reference Implementation

AetherMesh

Decentralized Peer-to-Peer Agent Synchronization and State CRDTs for Autonomous Multi-Agent Systems

A formal architecture for epidemic gossip propagation and lightweight Byzantine fault tolerance at agent scale. Now with a reference implementation (v2.0.0).

4
Formal Definitions
3
Theorems
3
Pillars
3
Experiments

Project

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

Central Question

How can autonomous agents synchronize shared code and metadata without central coordination — combining concurrent changes, disseminating them, and deciding which received changes to trust — and under which explicit assumptions may the combined guarantees hold?

Formal Definitions

Figure 8 — Gossip Propagation

Figure 8 — (a) Analytic curves of the gossip infection probability for β = 0.3, 0.5, 0.8. (b) Convergence latency versus network size for the five simulation-reported points.

DefinitionStatementStatus
1 — Convergent State TransitionR_a(t+1) = R_a(t) ⊔ ⨆_{m ∈ M_a} mDefined
2 — CRDT State Space(S, ⊔, ⊥) bounded join-semilattice; s_reconciled = s_i ⊔ s_jDefined
3 — Gossip Infection ProbabilityP_uninfected(t) ≈ e^(−β·t)Defined
4 — Byzantine Admission RuleLocal verification of provenance, structure, and semantics before admissionDefined

Figure 1 — Tripartite Architecture

Figure 1 — The tripartite architecture of AetherMesh: CRDTs, gossip, and lightweight BFT sit above the peer-to-peer agent network.

Theorems

Figure 2 — The Aether Family, Four Layers

Figure 2 — The Aether family, bottom to top: storage, integration, verification, decentralization. Each layer is shown with its DOI and the question it addresses.

#TheoremStatus
T1Eventual Consistency (under A1–A6)Proof sketch
T2Exponential Dampening of Gossip Latency (under A2)Proof sketch
T3Byzantine Fault Tolerance Bound (under A5, A6)Proof sketch
All three theorems are stated with proof sketches; none is machine-checked. They hold only under assumptions A1–A6.

Architectural Pillars

PillarStatus
1 — CRDTs for Code and Metadata DAGsImplemented
2 — Epidemic Gossip ProtocolsImplemented
3 — Lightweight Byzantine Fault ToleranceImplemented

Figure 4 — CRDT Join-Semilattice

Figure 4 — A seven-element bounded join-semilattice as a Hasse diagram. The join laws were verified on this example by exhaustive enumeration.

Figure 7 — Byzantine Admission Pipeline

Figure 7 — A remote delta passes through signature, DAG-integrity, schema, and semantic-compatibility checks before joining the local semilattice.

Empirical Results (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. The v1.0.0 paper remains the authoritative architectural record.

Installation

pip install aethermesh==2.0.0

Requires Python 3.9+. Runtime dependencies: numpy >= 1.24, cryptography >= 41.0.

Repositories and Archives

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.