Aethon

Instant agent instantiation through reference-based replication.

Aethon reframes stateful agents as compositional execution identities, making them cheap to spawn, safe to specialize, and practical to operate at production scale.

O(1)
instantiation goal with respect to inherited agent structure
10 pp
reference paper on stateful agent replication
3 layers
definition, reference, and resolution substrates
2026
arXiv paper on runtime primitives for agent infrastructure

The problem: stateful agents are too expensive to spawn

Modern AI agents are expected to preserve continuity, accumulate context, invoke tools, collaborate with other components, and remain operable inside workflows. But many runtimes still create instances by loading configuration, binding tools, reconstructing memory, and assembling executable objects from scratch.

The Aethon idea

Aethon treats an agent instance as a compositional view over a stable definition, layered memory, and local contextual overlays. Instead of duplicating inherited structure, a new instance can be represented as a lightweight reference record whose behavior is resolved when execution actually needs it.

Why this changes agent architecture

When instantiation is cheap, teams can create one-instance-per-user, one-instance-per-task, or dynamic multi-agent branches without forcing prompts to become overloaded monoliths. Specialization becomes a natural runtime operation rather than an expensive cold-start tradeoff.

What it enables for Next Moca

Aethon gives Next Moca a systems primitive for fast, governable agent identity: versioned definitions, shared memory layers, local overlays, lineage, scope, and policy can be composed without flattening everything into each runtime object.

Definition substrate

Stable role identity, behavioral instructions, declared capabilities, policy constraints, interface contracts, and structural metadata become the shared anchor for descendants.

Reference substrate

An instance is represented by identity, lineage, inherited pointers, and local deltas instead of a deep copy of every inherited structure.

Resolution substrate

At execution time, the resolver composes definitions, inherited memory, overlays, and scope constraints into the effective view the agent actually uses.