opensona
v0.2.1
Published
Toolkit for building RAG pipelines using Fandom wiki data, used to enhance LLM personas. Ships a Node CLI (download, prebuild, build, verify) and a browser + Node runtime (graph traversal with timeline-aware filtering).
Maintainers
Readme
opensona
A modular RAG (Retrieval-Augmented Generation) toolkit that grounds LLM personas in Fandom wiki content.
Built as a pnpm workspace package with two export paths: a CLI for building RAG bundles (Node-only) and a runtime for querying them (browser + Node).
Currently validated against the Cyberpunk 2077 Fandom wiki for MikoshiDB.net, but designed to work with any Fandom wiki with changes to the open source source code.
Installation
npm install opensonaRequires Node 22+. The opensona CLI is available via npx opensona after install.
Building (CLI)
Transform a Fandom wiki XML dump into a query-ready RAG bundle:
npx opensona download --wiki cyberpunk
npx opensona prebuild --config <path> --output <dir>
npx opensona build --config <path> --output <dir>
npx opensona verify --cases <path> --bundle <dir>Querying (runtime)
opensona runs a graph traversal; the caller owns the LLM round-trip that turns a user query into a TraversalDirective.
import {
createRuntime,
assembleLorePreamble,
buildResolverMessages,
parseTraversalDirective,
} from "opensona/runtime";
const rt = createRuntime();
await rt.load("/rag/");
const characterContext = {
id: "v",
bio: "Mercenary out of Night City.",
cutoffEventId: "arasaka-tower-raid",
};
const chunks = await rt.query("who is Adam Smasher", {
characterContext,
getTraversalPath: async (input) => {
const messages = buildResolverMessages(input);
const raw = await yourLLM(messages); // any chat completion that returns JSON
return parseTraversalDirective(raw);
},
});
const { source, license } = rt.manifest()!;
const preamble = assembleLorePreamble(chunks, { source, license });Configuration
opensona ships generic defaults in config.default.json. Consumers provide a JSON file via --config to override any field.
All wiki-specific data (source attribution, edition-era prefixes, category skip rules) lives in consumer config.
Example configuration
You can find opensona configuration for mikoshidb.net (cyberpunk.fandom.com) on GitHub
