Semantic canon engine

Field Horizon

Field Horizon is not a chatbot that improvises lore. It is a machine for turning fragments into worlds, while preserving provenance, contradiction, rejection and memory.

PythonRustSQLite FTS5PostgreSQL / pgvectorOllamaFastAPILocal web UI
Status
Active prototype
Code
FH-01
Domain
Semantic canon engine
Owner
Pomegranate Interactive

An AI-powered system that retrieves, interprets, disputes, verifies and accumulates complex fictional, theological and philosophical canon.

LocalCore inference
SemanticRetrieval model
AuditableCanon decisions

Mission profile

  • Ingest sacred texts, literary corpora, axioms and project documents.
  • Retrieve semantically relevant evidence instead of matching keywords alone.
  • Orchestrate specialized agents such as archivist, exegete, verifier and heresy hunter.
  • Accumulate canon only when claims survive evidence, contradiction and scoring.
  • Export structured lore, doctrine, world state and machine-readable artifacts.

Architecture

Why it exists

Most generative systems produce text faster than they produce meaning. Field Horizon reverses that priority. It treats canon as a contested institutional memory rather than a stream of plausible sentences.

What makes it different

Every output belongs to a traceable cycle. Sources are retrieved, claims are argued, truncation and uncertainty are detected, verdicts are recorded, and only sufficiently defensible fragments become canon. The machine is allowed to reject, defer or remain silent.

Current direction

The prototype already uses Python, SQLite FTS5, local models and a growing corpus. The next frontier is a generalized semantic layer, PostgreSQL and pgvector as optional scale-out infrastructure, a proper local GUI, and stable APIs for game engines and external tools.