add-memory

name	add-memory
description	Scaffold a new memory storage backend for spaceduck

Add Memory Backend

Scaffolds a new memory storage backend under packages/memory/<name>/.

Steps

1. Ask for the backend name (e.g., "pinecone", "pgvector", "milvus", "turbopuffer")
2. Determine if the backend needs an EmbeddingProvider (vector DBs do, text-search backends don't)
3. Create the directory structure:

packages/memory/<name>/
  package.json          # @spaceduck/memory-<name>, depends on @spaceduck/core
  src/
    store.ts            # <Name>ConversationStore implements ConversationStore
    long-term.ts        # <Name>LongTermMemory implements LongTermMemory
    index.ts            # barrel export
    __tests__/
      store.test.ts
      long-term.test.ts

4. Implement the ConversationStore interface:

import type { ConversationStore, Conversation, Message, Result, MemoryError, Lifecycle } from "@spaceduck/core";

export class <Name>ConversationStore implements ConversationStore, Lifecycle {
  // CRUD for conversations and messages
}

5. Implement the LongTermMemory interface:

import type { LongTermMemory, Fact, Lifecycle } from "@spaceduck/core";

export class <Name>LongTermMemory implements LongTermMemory, Lifecycle {
  // For vector backends: accept EmbeddingProvider via constructor
  // recall(query) internally embeds the query and does similarity search
}

6. Wire the backend in @spaceduck/gateway as an alternative to the default SQLite backend
7. Write integration tests using real DB connections (or in-memory variants)