CortexDB

Install & Quick Start

go get github.com/liliang-cn/cortexdb/v2

db, _ := cortexdb.Open(cortexdb.DefaultConfig("KnowledgeMemory.db"))
defer db.Close()

q := db.Quick()
_, _ = q.Add(ctx, []float32{0.1, 0.2, 0.9}, "SQLite is a single-file database.")
hits, _ := q.Search(ctx, []float32{0.1, 0.2, 0.8}, 1)

// No-embedder RAG (lexical):
_, _ = db.SaveKnowledge(ctx, cortexdb.KnowledgeSaveRequest{
    KnowledgeID: "apollo", Content: "Alice owns Apollo. Apollo ships Friday."})
resp, _ := db.SearchKnowledge(ctx, cortexdb.KnowledgeSearchRequest{
    Query: "Who owns Apollo?", RetrievalMode: cortexdb.RetrievalModeLexical, TopK: 3})

Examples & Status

examples/01_core … 15_cortex_query are small and architecture-oriented (go run ./examples/01_core); 01-07/09/15 run standalone, others need an LLM/embeddings/live DB — see examples/README.md.

An embedded local-first AI memory/KG library — not a drop-in replacement for Fuseki/GraphDB/Stardog. One file, Go APIs, tool/MCP surfaces, and enough RDF/SPARQL/RDFS/SHACL to build real memory workflows.

Tools, MCP & Plugin

tools := db.GraphRAGTools()                             // in-process tool calling
server := db.NewMCPServer(cortexdb.MCPServerOptions{})  // MCP server

Tool groups: GraphRAG (ingest_document, search_text, build_context), knowledge/memory (knowledge_save, memory_search, …), KG (knowledge_graph_query, _shacl_validate), KnowledgeMemory (knowledge_memory_recall, _reflect). memoryflow/graphflow/importflow/connector expose their own toolboxes too.

Claude Code and Codex plugin

Give Claude Code (and Codex) durable memory + a knowledge graph as a plugin. It bundles the cortexdb skill plus a live MCP server, runs in no-embedder lexical mode by default (no API key, no Go toolchain — the server binary is fetched from the matching release), and stores everything in one global SQLite file shared by every project.

Install — Claude Code — run each as a slash command:

/plugin marketplace add liliang-cn/cortexdb
/plugin install cortexdb@cortexdb
/reload-plugins

Install — Codex — run in your shell:

codex plugin marketplace add liliang-cn/cortexdb
codex plugin add cortexdb@cortexdb

Codex uses the same default global brain at ~/.cortexdb/cortexdb.db.

Use — just talk to Claude; it calls the MCP tools for you ("remember that I prefer …", "what do you know about X?"). Or use the slash commands: /remember <text>, /recall <query>, /cortexdb-graph (interactive knowledge-graph view), or /cortexdb for the skill. Key tools: memory_save / memory_search, knowledge_save / knowledge_search, knowledge_graph_query, and the unified knowledge_memory_recall. When enabled, a SessionStart directive + UserPromptSubmit auto-recall hook make Claude recall and save proactively (it asks once, per machine).

Where data lives — ~/.cortexdb/cortexdb.db by default, so memory follows you across projects (multiple sessions share it safely via SQLite WAL). Override per project:

export CORTEXDB_PATH=.cortexdb/cortexdb.db   # inherited by the launched server

To force the same global brain explicitly, set:

export CORTEXDB_PATH="$HOME/.cortexdb/cortexdb.db"

To upgrade: /plugin update cortexdb then /reload-plugins — the server binary auto-refreshes (version-pinned cache). See plugins/cortexdb/README.md for all env vars.