Skip to main content

Embeddings

Embeddings power semantic and hybrid retrieval. In the default Docker setup, Remembr uses Ollama with nomic-embed-text.

Lifecycle​

  1. POST /memory stores the episode immediately.
  2. The response includes embedding_status.
  3. A worker generates embeddings asynchronously.
  4. Semantic and hybrid retrieval improve once the status becomes ready.

Why pending matters​

If you store and search in the same request cycle, a just-stored episode may not appear in semantic results yet. Keyword search can still surface it if the text matches.

Providers​

  • Ollama by default for local and self-hosted setups
  • Other providers can be configured at deployment time