Building PlotIt: A High-Fidelity World-Building Engine with Agentic RAG

The Vision: More Than a Wiki

Creative writers often struggle with the "weight" of their own lore. As worlds grow, names, dates, and lineage relationships become a cognitive burden. PlotIt was born to solve this: it’s not just a wiki; it’s an intelligent "bible" for creative projects.

PlotIt centralizes complex fictional data—Characters, Races, Regions, Religions, and even entire Universes—into a structured, queryable graph, augmented by an agentic AI chronicler named Demitrei.

The Technical Core: A Layered Entity Architecture

To handle the scale of a world like Noor (the primary setting in our demo), I implemented a Layered Entity Pattern in the backend.

Why this matters

Most CRUD apps start simple but collapse under the weight of business logic. PlotIt uses a strict separation of concerns across every entity (e.g., character, universe):

Schema (character.schema.ts): Define the data shape using Drizzle ORM for PostgreSQL.
Repository (character.repository.ts): Pure database interactions.
Service (character.service.ts): Business logic, lore validation, and entity relationship management.
Controller & Router: RESTful endpoints with Zod for runtime validation.

Automation with Tether CLI

Building 13+ entities with 8 files each (Schema, Service, Router, etc.) is a boilerplate nightmare. To scale this, I built Tether, a custom CLI written in Rust.

Tether allows me to scaffold an entire backend/frontend vertical for a new lore entity in seconds: tether --name=religion --server. This ensures architectural consistency across the entire codebase.

Intelligence: Meet Demitrei

The standout feature of PlotIt is Demitrei, an agentic AI assistant designed to be a "Chronicler" of the user's lore.

The RAG Pipeline

Demitrei isn't just a chatbot; it's a Retrieval-Augmented Generation (RAG) system powered by:

FastAPI (Python) for the AI microservice.
Ollama (Qwen) for local LLM generation.
ChromaDB + Nomic for vector embeddings.

When a writer asks, "Who sired the first Arboreal in Noor?", Demitrei performs a semantic search across the Lore database, retrieves the relevant "fragments," and synthesizes an answer.

Multi-Channel Streaming & "Thinking"

To make the AI feel alive and trustworthy, I implemented a Multi-channel SSE (Server-Sent Events) stream.

Demitrei doesn't just output text; it streams:

Pulse: A heartbeat to show the engine is active.
Thinking: A "Chain of Thought" (CoT) block where the AI reasons through the lore.
Context: The exact lore documents retrieved from the vector store.
Tool Calls: When Demitrei needs to query the PlotIt API directly (e.g., to fetch a character’s family tree).

The Immersive Frontend

The UI was built to feel like an ancient terminal meets modern productivity tool.

Vite + React: For a lightning-fast dev experience.
Tailwind CSS + Framer Motion: For the "lore-friendly" aesthetic—smooth transitions, shimmer effects, and high-contrast typography.
Tiptap: A powerful headless editor for long-form lore writing.
XYFlow (React Flow): Used for visualizing complex relationships between characters and regions.

PlotIt represents a bridge between traditional world-building and modern AI-assisted creation. By combining a strict, relational PostgreSQL backbone with a flexible, agentic AI layer, we’ve created a tool that doesn't just store lore—it understands it.

Tech Summary:

Frontend: React, Vite, Tailwind, Framer Motion, Radix UI.
Backend: Node.js, Express, Drizzle ORM, PostgreSQL, Zod.
AI: Python, FastAPI, Ollama, ChromaDB, RAG.
DevTools: Tether CLI (Rust).

kdotdev