Ecosystem
Core LLM Wiki is the engine behind a family of products from Equational Applications. They all follow the same wiki pattern: events and documents go in, the librarian curates facts and a graph, and OKF bundles carry knowledge between products. They run on different platforms.
┌───────────────────────── Core LLM Wiki packages (this repo) ─────────────────────────┐ │ core-llm-wiki · react-llm-wiki · expo-llm-wiki · core-okf · ontologies · tools │ └──────────────┬──────────────────────────┬───────────────────────────┬────────────────┘ │ │ │ Curated Thoughts Curated Journal SynapseTree desktop · Tauri + React + Rust mobile · Expo, on-device cloud memory service │ │ curated-thoughts-integrations SynapseTree MCP server Claude Code · OpenCode · Hermes · …Curated Thoughts: desktop second brain
Section titled “Curated Thoughts: desktop second brain”Curated Thoughts is a desktop app
built with Tauri, React, and Rust. Its React front end runs react-llm-wiki, and a Rust
engine called the Active Librarian follows the same wiki patterns natively: it watches a vault
of PDFs, DOCX, and Markdown, chunks and embeds them locally, and proposes wiki pages.
It shows several Core LLM Wiki features in a real product:
- Human-in-the-loop review. Librarian proposals wait in a Review desk until you approve them. This is the draft-review idea used as a product workflow.
- Ontologies and a classifier. It ships the schema.org and software-organization ontology packages. It can also type facts with a Jev classifier through classifier-mode backfill.
- OKF backups. The whole brain exports to, and imports from, OKF v0.2 zip bundles.
- An MCP server. The app binary runs as a stdio MCP server (
--mcp), so Claude Desktop, Cursor, and other MCP clients can search, traverse, and add to your local brain.
A software agency in a box
Section titled “A software agency in a box”curated-thoughts-integrations connects that local brain to AI agent harnesses: Claude Code, OpenCode, Hermes Agent, DeepSeek Harness, OpenClaw, and Docker Agent. Every integration ships the same layers:
- MCP registration
- An install doctor
- Skills native to the harness
- Intuitive Wisdom: relevant curated facts arrive in the agent’s context at session start, on each user turn, and during autonomous work, never more than once per fact.
The result is a team of agents that share one curated, human-reviewed memory. A brain exported on one machine imports into any harness.
Curated Journal: mobile, fully offline
Section titled “Curated Journal: mobile, fully offline”Curated Journal is the flagship
reference app for expo-llm-wiki. It runs completely offline on
iOS and Android, and doubles as a cookbook for on-device memory:
- On-device inference. It runs GGUF models with
llama.rnthrough a singlegenerateText. - Night Shift. Auto maintenance is off. The librarian and heal run in the foreground, only while the device is charging. See Maintenance → on mobile.
- Emergent ontology. The model invents node and edge types, which the app shows in a Skia and d3-force graph explorer.
- OKF zip import and export, using the same format as Curated Thoughts.
Syncing with SynapseTree is on its roadmap, as an explicit, opt-in step.
SynapseTree: cloud memory
Section titled “SynapseTree: cloud memory”SynapseTree is the hosted member of the family. It runs Core LLM Wiki per tenant, with one SQLite file per tenant in object storage. The serverless hosting guide describes that pattern.
Which one fits?
Section titled “Which one fits?”| You want… | Use |
|---|---|
| Memory inside your own app | The Core LLM Wiki packages |
| A personal desktop brain, private and local | Curated Thoughts |
| One reviewed memory shared by your coding agents | Curated Thoughts + integrations |
| A private journal on your phone | Curated Journal |
| Hosted memory with no infrastructure to run | SynapseTree |