HOW-TO · DEV
How to use Continue.dev codebase index for fast semantic search across your project
Target environment
Ubuntu 24.04 · Continue.dev 0.9.x
PREREQUISITES
Continue.dev installed, project indexed
What this does
Continue.dev builds a semantic index of the local codebase using embeddings, enabling natural-language queries that return relevant code locations even when the search terms do not appear verbatim in the source. This transforms a large project into a navigable knowledge base where developers can ask questions such as "Where is authentication handled?" and receive precise file references with context snippets.
Steps
- Open the project workspace in VS Code.
- Locate the Continue.dev sidebar panel on the left side of the editor.
- Click the Index icon (a database symbol) in the Continue.dev panel header.
- Confirm the index status: it should show "Indexing in progress" and then "Index complete" once all files have been processed.
- Return to the chat input and type a semantic query, for example: "Where is the HTTP request handler defined?"
- Review the returned results: each entry shows a file path, a relevance score, and a snippet of surrounding code.
- Click any result to open the file at the indicated location.
- Re-run the index after adding or renaming many files by clicking Refresh Index to keep results accurate.
Verification
ls -la ~/.continue/indexes/$(basename "$(pwd)")/ 2>/dev/null && echo "Index directory exists and contains files"
Expected output: confirmation that the index directory for the current project was created and contains at least one index artifact file.
Common failures
- Index stuck at zero files: Large
node_modulesor binary directories may be excluded; add the project root to Continue.dev'sallowlist inconfig.jsonif it is outside the default scope. - Slow embeddings: Using a local Ollama embeddings model (such as
nomic-embed-text) significantly accelerates indexing compared to remote API calls. - Stale results after refactoring: Click Refresh Index after bulk file changes; otherwise the semantic search returns references to old code that no longer exists.