Nessy is an Autonomous AGI Intelligence Fabric. It is not a single monolithic model, but a modular framework that composes distinct computational capabilities into an integrated system. The architecture is designed to operate across distributed hardware, allowing local instances to run and federate without reliance on centralized data centers.
The system is built on two primary capability crates:
Planning Crate: Implements task decomposition, hierarchical organization, long-horizon plan generation, dynamic replanning, and resource scheduling.
Reasoning Crate: Implements structured logical inference, formal verification (pre/post-conditions), internal state reflection, hypothesis generation, and theorem proving.
These crates are bound to the Infinite Tree topology, which maps inference as a directed acyclic graph rather than a linear sequence. The Rivers Metric governs termination, replacing fixed token limits with a confidence-aware mathematical threshold. The system also implements an autonomous substrate (via .aider loops) that allows it to manage its own development cycles, testing, and integration.
Nessy operates on the Infinite Tree topology and uses the Rivers Metric for dynamic termination. It is designed for distributed federation, allowing local communities to host the AGI without relying on centralized data centers.