@baselineos/memory — Persistent Memory & Recall

Long-term memory for AI agents: episodic, semantic, and procedural recall with vector search.

What It Does

The Memory layer gives agents persistent, searchable memory across sessions:

• Episodic memory — Past tasks, decisions, and outcomes
• Semantic memory — Facts, relationships, and domain knowledge
• Procedural memory — Workflows, playbooks, and successful patterns
• Vector search — Embedding-based similarity retrieval (ChromaDB)

API

import { MemoryStore, type MemoryEntry } from '@baselineos/memory';

const store = new MemoryStore({ persistPath: '.baseline/memory' });
await store.initialize();

// Remember
store.remember({ key: 'customer-preference', value: 'formal-tone', scope: 'session' });

// Recall
const entries = store.recall('customer preference');

Package Boundary

This package owns:

• Memory entry storage and retrieval
• Vector embedding and similarity search
• Scope-based memory isolation (session, task, agent, global)
• Memory compaction and eviction

It does not own:

• Knowledge indexing (Indexer layer)
• Credential storage (Vault layer)
• Audit trails (Govern layer)

License

Apache-2.0