Cursor (Anysphere)Cursor retrieval
Cursor Code Embedding Model
Cursor trains a retrieval model from agent sessions. An LLM ranks which code would have helped at each step, and the embedding model learns those rankings.
What the release provides
The model connects natural-language questions with relevant code segments. It is used with indexing and text search in Cursor. The report does not publish a standalone model ID, weights, vector dimensions or public embedding API.Cursor’s custom code embedding model
How to use it well
Practical guidance. Inspect retrieved code in its surrounding module before drawing a conclusion. Combine semantic queries with exact symbol searches when a missed call site could change the answer.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Provider product |