@snoai/embedder

Local text embeddings for Sno Station: the part that lets memory be found by meaning, with no API key and no network call after the first download.

It runs a quantized ONNX embedding model on your machine (1024-dimension vectors, pinned to a fixed model revision) through @huggingface/transformers, and adds an in-memory LRU cache so repeated text is not embedded twice.

• LocalEmbedProvider creates the embeddings; CachedEmbeddingProvider wraps it with the cache.
• ensureModelDownloaded fetches the model once into a cache folder, and isModelCached tells you whether it is already there. The memory service setup calls it, so your first agent session does not wait for the download.

import { ensureModelDownloaded, LocalEmbedProvider } from "@snoai/embedder";

Who it is for: the memory engine, and anyone building on it. If you only want agent memory, you install it as part of the memory service setup, then add the plugin for your agent (Codex, Claude Code, OpenClaw).

npm install @snoai/[email protected]

Licensed under Apache-2.0.