@rbrtdds/acp-embeddings

Local embedding provider for ACP (AI Context Protocol) — offline semantic search using transformers.js.

Install

npm i @rbrtdds/acp-embeddings

This is an internal dependency of @rbrtdds/acp-cli and @rbrtdds/acp-mcp. You don't need to install it directly unless you're building custom integrations.

What it does

Provides a local embedding provider that generates 384-dimensional vectors using the all-MiniLM-L6-v2 model via transformers.js. The model (~23MB) is downloaded automatically on first use and cached in ~/.acp/models/.

• Fully offline — no API keys or network required after initial download
• Used by ACP's recall engine for semantic (hybrid) search
• Falls back gracefully — ACP works with keyword-only search if embeddings are disabled

Usage

import { LocalEmbeddingProvider } from '@rbrtdds/acp-embeddings';

const provider = new LocalEmbeddingProvider();
await provider.initialize();

const vector = await provider.embed('authentication middleware');
// Float32Array(384) [0.023, -0.041, ...]

Configuration

Enabled/disabled during acp init. Stored in ~/.acp/config.json:

{
  "embedding": {
    "engine": "local",
    "model": "Xenova/all-MiniLM-L6-v2",
    "dimensions": 384
  }
}

Set engine to "none" to disable embeddings (keyword-only search).

Related

• @rbrtdds/acp-core — Core library
• @rbrtdds/acp-cli — CLI tool
• @rbrtdds/acp-mcp — MCP server for Claude Code

License

MIT