@eidentic/memory

Self-improving memory engine for Eidentic — four-tier recall with lexical + semantic RRF fusion, self-editing memory blocks, a temporal knowledge graph, passive fact extraction, and sleep-time consolidation. This package provides the Memory class that wires a StorePort and VectorPort together into Eidentic's full memory stack.

Install

pnpm add @eidentic/memory

Usage

import { Memory } from "@eidentic/memory";
import { SqliteStore } from "@eidentic/sqlite";
import { LanceDBVectorStore } from "@eidentic/lancedb";
import { AIEmbedder } from "@eidentic/model";
import { openai } from "@ai-sdk/openai";

const store = new SqliteStore("./eidentic.sqlite");
const embedder = await AIEmbedder.create(openai.embedding("text-embedding-3-small"));
const vector = await LanceDBVectorStore.open("./lancedb", "memory_vectors", 1536);

const memory = new Memory({ store, vector, embedder });

// Retrieve relevant snippets for a session
const scope = { kind: "user" as const, agentId: "my-agent", userId: "u-1" };
const result = await memory.retrieve({ text: "budget decisions", scope, topK: 5 });
console.log(result.snippets);

Links

• GitHub
• Issue tracker
• Root README

Apache-2.0