LanceDB Vector Database Management in VS Code

Overview

LanceDB is an open-source, embedded (in-process) vector database built on the Lance columnar data format. Highlights include:

• In-process, no server: A database is just a directory of Lance datasets on disk - no service to run, no credentials
• Built on the Lance format: A fast, versioned columnar format designed for ML and vector workloads
• Strict schema: Each table declares its columns (an Arrow schema), including the vector column (a fixed-size float list)
• SQL filtering: Filter searches and browses with SQL expressions (evaluated by DataFusion)
• Native performance: A prebuilt platform binary, downloaded on first connect

LanceDB is commonly used for local-first semantic search, retrieval-augmented generation (RAG), and embedding-heavy applications that want a database file rather than a server.

Connecting

To connect to LanceDB in DBCode:

1. Open the DBCode Extension: Launch Visual Studio Code and open the DBCode extension.
2. Add a New Connection: Click on the “Add Connection” icon.
3. Complete the connection form: Select LanceDB as the database type and choose the Database Directory - the folder containing your Lance datasets. There is no host, port, or credential.
4. Connect: Click save. The first connection downloads the platform binary for your OS.
5. Start exploring: Browse your tables, inspect the schema, and run vector searches.

For detailed instructions, refer to the Connect article.

LanceDB Features in DBCode

DBCode brings the same browse-and-search workflow you already use for SQL and document databases to LanceDB:

• Table browsing: Navigate tables and inspect the declared Arrow schema (id, vector, and scalar columns)
• Vector cell rendering: Vector columns are summarised inline (e.g. [float32×768]) and expandable on click
• Vector search: Run nearest-neighbour searches with top-K, SQL filters, and a _score column (Lance returns a distance, so lower is closer)
• Editing: Edit scalar columns inline and delete rows (the id and vector are read-only)
• Search by text: Configure an Ollama model or DBCode AI to embed your query text on the fly (LanceDB has no built-in embedding)
• JS shell editor: Drop into a JavaScript editor and run the LanceDB client directly (client.search('table', { vector, limit, where }), client.browse(...), table('name')...)

By using LanceDB with DBCode, you get a unified workspace for traditional and vector data without leaving VS Code.

For more information about LanceDB, check out LanceDB.