Dashboard
Overview of your semantic search setup
⌘K
Your home base. Stats show tables configured, total embeddings, and pending work. Click any table row to search it. Use Start Worker to process pending rows. Press ⌘K for the command palette.
Configured Tables
Loading...
Worker Health
Loading...
Search
Natural language semantic search
Search using natural language — describe what you're looking for and pgsemantic finds rows with similar meaning. Add filters to narrow by column values (e.g. category, author). Similarity scores are relative — higher is better, but exact thresholds depend on your data and model. Try a few queries to calibrate.
Inspect Database
Discover columns suitable for semantic search
Select a table and click Scan to analyze its columns. Columns are scored by text content quality — longer, meaningful text scores higher. Click Apply on any column to set up semantic search.
Apply
Set up semantic search on a table
Creates an embedding column (or shadow table), HNSW index, and change-tracking trigger. Model:
Local runs on your machine (no API key). OpenAI needs OPENAI_API_KEY. Ollama needs Ollama running locally. Shadow table stores embeddings separately without modifying the original table.OpenAI API Key
Required for OpenAI Small and OpenAI Large above. Saved to your
.env file — never exposed to the browser.Index
Bulk embed existing rows
1
Configure
Pick the table to embed and how many rows to process at once.
2
Embedding Progress
Live status while the model generates vectors for your rows.
Start indexing above to see progress here.
3
Worker — Auto-embed New Rows
New rows inserted into your table are queued by a trigger. The Worker runs in the background and embeds them automatically so search stays up to date.
Checking...
Visualize
2D embedding scatter plot
See how your data clusters semantically. Each dot is an embedded row — hover to see its text, click for details. Points that are close together are semantically similar.
Status
Embedding health dashboard
Coverage = percentage of rows with embeddings. Pending = rows waiting to embed. Failed = rows that couldn't embed (empty content, encoding issues). Enable Auto-refresh to watch progress live.
Loading...
Worker Health
Loading...
Model Migration
Table Browser
Explore your database tables
Browse all tables in your database. Click any row to see columns, types, and sample data in a table view. Use Inspect to evaluate a table for semantic search.
Loading...
SQL Query
Run read-only queries against your database
Execute SELECT queries directly. Useful for exploring data, checking counts, and running analytics. All queries are read-only with a 10-second timeout and 100-row limit. No mutations possible.
MCP Integration
Connect AI agents to your semantic search
MCP lets AI assistants (Claude Desktop, Cursor) search your database directly. stdio: Claude runs pgsemantic as a subprocess (recommended). SSE: connects to this server's endpoint. Click Configure to auto-write the config, then restart your AI client.
Server Status
Claude Desktop
Config Snippets
Click to copyLoading...
Loading...
Available Tools
Connection
Database connection setup
Connect to any PostgreSQL database with pgvector. Works with localhost, Docker, AWS RDS, Supabase, Neon — any PostgreSQL host. Format:
postgresql://user:pass@host:5432/db. Test checks connectivity without saving. Save writes to your .env file.