Bring your own inference
You give it one generateText function. It builds the prompts, parses the JSON, and writes to
the database. It works with OpenAI, Gemini, Claude, or a local model.
Core LLM Wiki
Long-term memory for your AI agents. Facts, tasks, and a knowledge graph, all stored in SQLite and running wherever your app runs.
Why Core LLM Wiki?
Section titled “Why Core LLM Wiki?”Just SQLite
Facts, vectors, and graph edges all live in tables in your database, with a
llm_wiki_ prefix by default. You don’t run a vector database or Neo4j.
Offline first
Keyword search runs in-process. Semantic search turns on when you provide embed(), and
falls back to keyword search when the network is down.
GraphRAG built in
Optional ontologies make the model emit typed edges. traverseGraph() then walks them
with a single recursive SQL query.
Pick your platform
Section titled “Pick your platform”npx expo install expo-sqlite expo-cryptonpm install @equationalapplications/expo-llm-wikiThe SQLite adapter, secure ID generation, and React hooks are already wired up. Expo quickstart →
npm install @equationalapplications/react-llm-wiki sql.jsRuns entirely in the browser on SQLite compiled to WebAssembly. React quickstart →
npm install @equationalapplications/core-llm-wiki better-sqlite3Use it on a backend or in a worker, with any SQLite driver. Node.js quickstart →
Three lines to memory
Section titled “Three lines to memory”await wiki.write('user-123', { event_type: 'observation', summary: 'Loves hiking on weekends.' });const memory = await wiki.read('user-123', 'what should we do on Saturday?');const context = formatContext(memory); // → paste into your system promptBuilt with Core LLM Wiki
Section titled “Built with Core LLM Wiki”Curated ThoughtsLocal-first desktop second brain (Tauri + Rust), with integrations that turn it into a software agency in a box for your AI agents.
Curated JournalFully offline Expo journal with on-device LLMs. The reference app for expo-llm-wiki.
SynapseTreeHosted cloud memory built on the same engine, with an MCP server for coding agents.
Where to next
Section titled “Where to next”How memory worksEvents, facts, tasks, and the librarian that connects them.
GuidesTask-focused how-tos: ingest documents, tune retrieval, add a graph.
RecipesComplete, copy-pasteable examples you can build from.
API referenceEvery method, option, hook, and error class.