Crate mw_memory 

mw-memory — explainable retrieval for MemoryWhale.

The differentiator: retrieval that returns not just which memories, but why — a blended score over interpretable signals (similarity, recency, importance, reinforcement, task-relevance), each with a human-readable reason.

Everything is behind a engine::MemoryEngine interface that MemoryWhale owns, so the storage/retrieval backend is pluggable: the built-in scorer by default, or — with the off-by-default mempalace feature — an engine::MemPalaceEngine that talks to a local mempalace-mcp server over MCP. Callers never change.

Modules

embed

Pluggable text embedding — so similarity can be semantic, not just lexical.

engine

The pluggable retrieval backend.

scorer

The explainable scorer: turn a (memory, query) pair into a blended score plus the interpretable signals behind it.

sqlite

The one loader: build scorable Memory items from a MemoryWhale SQLite DB.

Structs

Memory

A single memory item. (The built-in engine scores over these; the MemPalace adapter maps its hits into the same shape.)

Query

A retrieval request plus the context that makes scoring explainable.

ScoredMemory

A memory with its blended score and the signals that produced it.

Signal

One interpretable scoring signal and its contribution.

Weights

How much each signal counts toward the blended score.