Epistemic Graph

epistemic-graph is the Rust-native compute engine for the agent-utilities ecosystem. It consolidates graph algorithms, quantitative finance, data science primitives, AST analysis, and OWL/Datalog reasoning into a single high-performance binary, exposed to Python out-of-process through a long-running Tokio daemon that speaks length-prefixed MessagePack over Unix Domain Sockets (or TCP) and authenticates with HMAC-SHA256.

This site collects the engineering references for operating and extending the engine. Use the navigation to move between the technical overview, the service-mode operations guide, the Rust compute reference, the measured transport benchmarks, and the concept registry.

Documentation map

• Technical Overview — the compiled Rust data structures and the graph algorithms (topological sort, cycle detection, shortest path, blast radius, PageRank, community detection) that the engine exposes.
• Service Mode — running the engine as a daemon: socket and TCP endpoints, authentication, sharding, and health management.
• Rust Compute Guide — the architecture of the compute modules and the procedure for adding a new capability across the protocol, server, and client layers.
• Transport Benchmarks — measured per-operation latency over the MessagePack transport, with reproduction instructions.
• Concept Registry — the stable CONCEPT identifiers that trace the engine's core ideas across code and documentation.

Design principle

Every invocation crosses a process boundary: serialize, socket round-trip, deserialize. A call is not a cheap function call. Batch work into a single round-trip over data already resident in the graph, and keep tight per-element math in-process — the Rust Compute Guide explains how this shapes every caller.