Skip to main content
MemWire is built around a set of distinct primitives that go beyond basic memory storage. Each feature is composable, you can use all of them together or only the ones your application needs.

Graph Memory

Token-level displacement graph connects memories through semantic relationships for multi-hop recall.

Multi-Tenancy

Native four-level hierarchy isolates memory across orgs, workspaces, apps, and users.

Knowledge Base

Ingest documents and chunks that are searched alongside episodic memories at recall time.

Hybrid Search

Dense and sparse (SPLADE) vectors combined for retrieval that catches both semantic and exact matches.

LLM-Free Classification

Every memory is auto-tagged by category using smart anchor similarity with zero external LLM API calls.

Adaptive Feedback Loop

Edge weights update from real LLM responses, so the memory graph improves with every conversation.