Engram Memory icon

Engram Memory

Knowledge graph memory for AI agents. Stores conversations as episodes and optionally extracts entities and relationships.

Engram Memory

AI-generated

Summary

Connects Engram as AI agent memory, storing conversation episodes and optionally extracting entities and relationships into a persistent knowledge graph. On later turns, retrieves relevant graph context for the agent, with optional filters, semantic search, traversal, and retention settings.

Inputs

  • Session ID — By default, reads sessionId from incoming data. Choose Define below and set Key (sessionKey) to provide a custom expression or static session key; this scopes the memory graph.
  • Storage Backend — Choose Embedded (Graphology), which persists graph files locally, or Neo4j (Remote). Embedded storage optionally accepts a Custom Storage Path; otherwise it uses the default n8n storage directory.
  • Context Window Length — Number of recent conversation turns to include as context; defaults to 10.
  • Knowledge Extraction — When enabled, uses the configured Extraction Model to extract entities and relationships from eligible conversation messages. Extraction requires an Engram Extraction LLM credential. Configure entity types, fact confidence thresholds and policy, and whether human, AI, or system/tool content is eligible.
  • Semantic Search (Embeddings) — When extraction is enabled, optionally select an Embedding Model to generate vectors and use hybrid text and vector search.
  • Graph Traversal (BFS) — Optionally expand retrieved context to related graph entities and facts; set Traversal Hops to control the search depth.
  • Retrieval filters — Optional token budget, maximum facts, minimum relevance score, fact policy, and filters for sender, episode kind, trust level, source workflow or execution, and reference time constrain the context retrieved for the agent.
  • Episode storage and metadata — For version 2 settings, choose whether to store human and AI episodes. Optional fields set human episode kind, message and idempotency IDs, conversation and sender details, quoted message ID, trust level, and JSON episode attributes.
  • Retention Policy — Choose to keep episodes forever, for a number of days, or up to a maximum episode count; Retention Value sets the days or count.

Output shape

An AI memory connection consumed by an AI agent; this operation does not return regular workflow items.

The node must be connected to an AI agent. Memory context is assembled from recent conversation turns and, when configured, relevant knowledge graph results.

Links

Discussion