Sandeep Chakravartty

Posted on Sep 30

Canon Keeper: Reconciling Troublesome Lore

#devchallenge #sanitychallenge #sanity #ai

Sanity Challenge Path One Submission

This is a submission for the Sanity Challenge, Path One: Ship an Agent That Queries Real Content

What I Built

CanonKeeper is an AI-powered Fictional Universe Lore & Canon Resolution Engine.

When building expansive fictional universes (fantasy novels, sci-fi series, tabletop RPGs, or gaming franchises), lore accumulated over years frequently suffers from retcons, timeline contradictions, bardic folklore myths, and multi-hop custody chains. Traditional keyword search fails when an early edition book claims a commander entered a citadel in 734, but a later primary garrison ledger signed by the commander proves he remained at a northern outpost until 735.

CanonKeeper solves this by storing structured lore documents in Sanity CMS and running an intelligent Lore Verification Agent that:

Queries atomic claims, source documents, characters, events, and locations via Sanity GROQ & MCP.
Applies Source Authority Hierarchy Weighting (PRIMARY_CANON > OFFICIAL_SECONDARY > REFERENCE > COMMUNITY).
Automatically detects and resolves Timeline Retcons (e.g., 1st Edition Chronicles vs. Imperial Garrison Ledgers & Revised Edition).
Performs Multi-Hop Evidence Deduction to trace artifact possession and event chains across dates.
Renders an Interactive SVG Evidence Graph mapping nodes (Question, Claim, Source, Verdict) and directed relationship edges (retrieves, supported_by, contradicts, supersedes, resolves_to).

Demo

Watch the complete video walkthrough of CanonKeeper in action:

Video Link: https://youtu.be/Qoc_rMhq9z4

Code

Check out the full open-source codebase on GitHub:

🐙 GitHub Repository: https://github.com/scha54/Sanity-MCP-Hackathon

How I Used Sanity

Sanity served as the structured graph foundation for CanonKeeper's knowledge base.

1. Structured Schema Modeling

I designed 5 custom Sanity document schemas:

claim: Represents atomic lore facts with subject-predicate-object triples, source reference, canonStatus (CANON, RETCONNED, DISPUTED, AMBIGUOUS), timeline markers, confidence scores, and reference arrays for contradicts, corroborates, and supersedes.
source: Documents with sourceType (CHRONICLE, MANUSCRIPT, BARDIC_SONG, LEDGER), edition, publicationDate, authorityLevel, and supersedes / supersededBy references.
character: Character entities with aliases, descriptions, first appearance references, and typed relationships.
event: Chronological events with location references, participants, and preceding/following event links.
location: World locations linked to associated characters and events.

2. Sanity Model Context Protocol (MCP) Integration

I configured the Sanity Model Context Protocol (MCP) server integration (https://mcp.sanity.io) in .agents/mcp_config.json. This enables AI agents to inspect schema types and query content via standardized MCP tool calls.

3. GROQ Evidence Queries & Resolution Logic

The agent retrieves evidence from Sanity using GROQ queries:

*[_type == "claim"]{
  _id,
  statement,
  subject,
  predicate,
  object,
  "sourceId": source._ref,
  sourceLocation,
  canonStatus,
  timeline,
  confidence,
  "contradictIds": contradicts[]._ref,
  "corroborateIds": corroborates[]._ref,
  "supersedeIds": supersedes[]._ref,
  notes
}

What the agent does with retrieved Sanity content:

Entity & Intent Recognition: Extracts entities (e.g., Arin Shadowstrider, Valemont Citadel) and timeframe constraints from user queries.
Graph Traversal: Fetches matched claims, resolves linked source document authority levels, and expands contradicts / supersedes graph nodes.
Authority & Retcon Resolution: Overrides outdated claims when a higher authority document (e.g., PRIMARY_CANON revised ledger) supersedes an earlier OFFICIAL_SECONDARY account.
Interactive Visualization: Constructs node-link network graphs displaying claims, sources, and canon verdicts.

Sanity Project Details

Sanity Project ID: demo-universe (Local fallback & Sanity dataset ready)
Dataset: production
Seeding Script: Included at scripts/seed-sanity.ts to populate Sanity CMS datasets using @sanity/client.

DEV Community