Semantic search helps you find related meaning even when your query does not repeat the document's exact words. ThinkForge prepares local representations of document text, called embeddings, to support this search and related-document suggestions.
Use semantic search
In the folder tree's Search tab, enable Smart and enter a subject or question. Files search also combines words with related meaning.
Search preparation and matching run on your computer using a model bundled with ThinkForge. No project text is sent to an online embedding service. This does not change how AI Tools or connected AI clients share the context you give them.
See Search and Retrieval for choosing the search control.
Wait after a large import
New and changed documents need to be processed before all semantic results are available. A large import can take time.
Search by a filename or path while processing continues. If a result is missing, also check that its file type and location are included in project search.
When semantic search is unavailable
There is no user-facing switch for local embeddings. They are available automatically when the bundled model is present.
Missing or damaged model files can prevent semantic search. Keyword and tag searches remain available; combined search can continue with word matches.
A connected MCP client can check whether embeddings are available and whether processing is complete. If the installed model is missing, repair or reinstall ThinkForge while preserving your projects and application data. See Installation and Updates.