Search models: embeddings and a reranker

Without a model, search is by words. System › Search models… (SPC s) adds the models that search by meaning. It has two tabs.

Embedding Models turn each section of your files into a vector, so a search finds what means the same.

1. Add… a provider: a local Ollama with an embedding model (ollama pull nomic-embed-text), Ollama Cloud, or any OpenAI-compatible endpoint with its key.
2. Use it. The workspace is indexed in the background; the status line's mark names the model searching.
3. Cancel indexing stops a long first pass; Words only goes back to search by words.

Ranker Models are a second stage: a reranker reads the top hits against your question and orders them again.

1. Add… a ranker: its kind, base URL and model; Check tries it.
2. Use it. Window… sets how many of the top hits it reads.

flowchart LR
  Q[Your search] --> W[By words]
  Q --> M[By meaning]
  W --> H[The top hits]
  M --> H
  H -->|a reranker| O[Ordered again]

The search's checkboxes (Lexical, Semantic, Rerank) choose the stages for each search.

More: Reference, Chapter 6: Configuring Embeddings and Rerankers.