Domain-agnostic knowledge graph memory for AI agents, v0.19.6. Persists structured entity-relationship graphs from conversations using local SQLite with FTS5. Supports zero-config offline mode with rule-based extraction, optional LLM-fallback for complex entities, natural language queries, and path finding. Ships TypeScript types. Active development with weekly releases. Key differentiators: fully local-first (no API keys needed), temporal facts with confidence decay, OpenClaw plugin integration, MCP server support, and relation normalization.
npm install agent-memory-graphNo compatibility data collected yet for this library.
Verified import paths — ran on the pinned version, not inferred.
Creates a local graph, ingests a sentence, and queries it by natural language.
new MemoryGraph({ dbPath: './memory.db' }) instead of new MemoryGraph('./memory.db')Remove reliance on the returned entity/relation count. Use graph.stats() if needed.
Check for null: const path = graph.findPath('A', 'B'); if (path) { ... }On Node 24, install gcc, make, python3, or use prebuilt binaries from better-sqlite3 releases.
Normalize casing before querying, or use findPath with regex.
Run 'npm rebuild better-sqlite3' or install build tools: on Ubuntu 'sudo apt install build-essential python3'
Use import { MemoryGraph } from 'agent-memory-graph'Wrap dbPath in options object: new MemoryGraph({ dbPath: './mydb.sqlite' })