@moxxy/plugin-embeddings-openai wraps OpenAI's embedding API as an EmbeddingProvider. Pair with @moxxy/plugin-memory to enable vector-based recall.

Install

pnpm add @moxxy/plugin-embeddings-openai

Use

import { createOpenAIEmbedder } from '@moxxy/plugin-embeddings-openai';
import { buildMemoryPlugin } from '@moxxy/plugin-memory';
import { CachedEmbeddingProvider } from '@moxxy/sdk';

const embedder = new CachedEmbeddingProvider({
  inner: createOpenAIEmbedder({
    apiKey: process.env.OPENAI_API_KEY!,
    model: 'text-embedding-3-small', // or 'text-embedding-3-large'
  }),
  cacheDir: '~/.moxxy/memory/.embeddings',
});

const { plugin } = buildMemoryPlugin({ embedder });
session.pluginHost.registerStatic(plugin);

memory_recall(mode: 'auto'|'vector') now produces embeddings via OpenAI.

Exports

• OpenAIEmbedder, createOpenAIEmbedder
• OpenAIEmbedderOptions, OpenAIEmbeddingModel
• CachedEmbeddingProvider — re-exported from @moxxy/sdk for back-compat; new code should import directly from the SDK.

Caching

CachedEmbeddingProvider writes one file per text-hash; the next call for the same input is a disk hit. Drop the cache dir to force regen.