embedcache

Use case

Air-gapped embeddings

Generate embeddings entirely on-box for environments where sending text to a third-party API is not an option.

The problem

Regulated and sensitive workloads cannot send document text — or its metadata — to a hosted embedding endpoint. But they still need vectors for search and retrieval.

How embedcache helps

  1. 1

    Install the embedcache service or crate on hardware inside your boundary.

  2. 2

    Bundled ONNX model weights mean no network call is needed to embed — inference runs locally on CPU.

  3. 3

    The SQLite cache is a single local file you control, back up, and can wipe on demand.

  4. 4

    Optionally skip LLM chunking (or point it at a local Ollama) so nothing crosses the boundary.

Features involved