Reference

Packages

Engram is published as a Turborepo monorepo. Each package is independently versioned and published to npm under the @engram scope.

@engram-ai-memory/mcp0.1.0

npm ↗

MCP server for Claude Code and any MCP-compatible client. Exposes 18 tools over the Model Context Protocol via stdio transport — memory, search, recall, decay, contradictions, tags, webhooks, and more.

npm i -g @engram-ai-memory/mcp
# or
pnpm add -g @engram-ai-memory/mcp

engram-mcpCLI binary — runs the MCP server (stdio transport)

store_memoryMCP tool — store a new memory

recall_contextMCP tool — assemble relevant context for AI prompts

search_memoryMCP tool — semantic vector search

add_knowledgeMCP tool — store a semantic fact

memory_statsMCP tool — get brain statistics

forgetMCP tool — archive memories

@engram-ai-memory/cli0.1.0

npm ↗

Terminal power-user interface. Store, search, recall, forget, export, and import memories from the command line. Uses @engram-ai-memory/core directly — no server required.

npm i -g @engram-ai-memory/cli
# or
pnpm add -g @engram-ai-memory/cli

engram storeStore a new memory with type, importance, tags

engram searchSemantic vector search

engram recallAssemble working memory context

engram statsMemory store statistics and health

engram forgetArchive a memory by ID

engram exportExport all memories as JSON/NDJSON

engram importImport memories from JSON/NDJSON

@engram-ai-memory/core0.1.0

npm ↗

Core TypeScript library. The brain — memory engine, embedding pipeline, vector search, knowledge graph, and context assembly. Embed Engram into your own Node.js application.

npm i @engram-ai-memory/core
# or
pnpm add @engram-ai-memory/core

NeuralBrainMain class — store/recall/search/forget/stats

Embeddertext → FP16 384-dim vector using ONNX (all-MiniLM-L6-v2)

VectorSearchIn-memory cosine similarity search index

KnowledgeGraphGraph engine with BFS expansion and 7 relationship types

ContextAssembler7-step recall algorithm — embed → search → expand → score → rank → truncate → log

ImportanceScorerComposite scoring: similarity + recency + importance + access frequency

@engram-ai-memory/vis0.1.0

npm ↗

The React Three Fiber 3D visualization dashboard. Can be embedded in any React 18+ app. Ships with 5 view modes and a full UI (memory list, inspector, status bar).

npm i @engram-ai-memory/vis
# peer deps: react react-dom three @react-three/fiber

NeuralCanvasFull dashboard component with sidebar and inspector

MemoryGraphBare 3D canvas — nodes + edges only

useMemoryStoreZustand store hook for memory state

useViewStoreZustand store hook for active view/theme

VIEWSArray of 5 view configs (Cosmos, Nebula, Neural Net, Galaxy, Clusters)

Versioning

All packages are versioned together — a release always updates all three packages to the same version. This guarantees API compatibility across the monorepo.

TypeScript support

All packages ship with full TypeScript declarations. No @types/ package is required. Engram uses exactOptionalPropertyTypes and strict mode throughout.

Node.js compatibility

Package	Node.js	Notes
@engram-ai-memory/mcp	22 +	Requires better-sqlite3 native addon (node-gyp)
@engram-ai-memory/core	22 +	@xenova/transformers ONNX runtime, better-sqlite3
@engram-ai-memory/vis	Browser	React 18+ component — no Node.js runtime required

← PreviousREST API