Database

The Database tab provides a built-in browser for your agent's data stores. Access it from the Advanced section of the dashboard at /database.

Overview

The database browser has three sub-tabs for different data types:

Tab	Description
Tables	Browse relational database tables — view rows, columns, and record counts
Media	Browse uploaded and generated media files (images, audio, video)
Vectors	Inspect vector store entries used for semantic search and RAG

Tables

The Tables view lists all database tables used by the agent runtime. Select a table to browse its contents in a paginated data grid.

Common tables include:

Table	Contents
memories	Agent memories and conversation state
messages	Conversation history
knowledge / documents	Knowledge base documents and fragments
entities	People, agents, and other entities the agent knows about
rooms	Conversation rooms and channels
tasks	Scheduled and active tasks
triggers	Event triggers and their run history

Select any table to view its rows in a paginated grid. Each row displays all columns with their values.

Media Gallery

The Media view scans database tables for embedded media URLs (images, videos, audio) and presents them in a filterable, searchable grid with a lightbox viewer.

How Media Is Discovered

The gallery scans up to 10 database tables, prioritizing tables whose names contain memor, message, media, attach, file, asset, or document. For each row, it extracts HTTP URLs and data: URIs from all string-valued columns, including JSON content blobs.

Supported Media Types

Type	Extensions	Badge Color
Image	.png .jpg .jpeg .gif .webp .svg .bmp .ico .avif	Blue
Video	.mp4 .webm .mov .avi .mkv .ogv	Purple
Audio	.mp3 .wav .ogg .flac .aac .m4a .opus	Green

Filtering

• Text search — filter by filename or URL (case-insensitive)
• Type filter chips — All / Images / Video / Audio (mutually exclusive)
• Results are deduplicated by URL and sorted by creation date (newest first)

Lightbox

Click any media item to open the lightbox:

• Images display at up to 70% viewport height
• Video plays with native controls
• Audio plays with native controls
• Footer shows type, source table, and creation date
• Close with click outside, Escape, or Enter

Vector Browser

The Vectors view browses agent memories and vector embeddings with three visualization modes. It auto-discovers vector-relevant tables (memories, embeddings, knowledge, vector) and joins embedding data when available.

Toolbar

• Table selector — choose which vector table to browse
• Stats bar — total count, embedding dimensions, unique memory count
• Search — filter by content text (SQL LIKE query)
• View mode toggle — List, Graph (2D), or 3D
• Refresh — reload data

Embedding Support

The viewer detects elizaOS embedding tables with dimension columns (dim_384, dim_512, dim_768, dim_1024, dim_1536, dim_3072). When the memories table is selected and an embeddings table exists, the viewer performs a LEFT JOIN to attach embedding vectors to memory records.

List View

Paginated cards (25 per page). Each card shows:

• Content preview (first ~200 characters)
• Type badge and Unique badge
• Room ID, Entity ID, Created At
• Embedding dimension count (e.g., "384 dims")

Click any card to open the Memory Detail Modal with full content, metadata grid, embedding values, and raw JSON record.

Graph View (2D Scatter Plot)

Projects high-dimensional embeddings to 2D using PCA (power iteration, 2 principal components). Renders to an HTML canvas.

• Axes labeled PC1 (horizontal) and PC2 (vertical)
• Points colored by memory type with a legend
• Hover — nearest point highlights with a tooltip showing content preview (60 chars)
• Click — opens the Memory Detail Modal
• Requires at least 2 records with embeddings (fetches up to 500)

3D View (Three.js)

Projects embeddings to 3 principal components and renders in an interactive Three.js scene.

• Each memory is a colored sphere positioned in 3D space
• Mouse drag — orbit camera around the scene (with damping)
• Scroll wheel — zoom in/out (radius clamped 2–15)
• Hover — highlighted sphere scales up with content tooltip (150 chars)
• Click — opens the Memory Detail Modal
• Scene includes a grid floor and 3-axis lines for spatial reference
• Points colored by memory type with a legend below the canvas

Memory Detail Modal

Full-screen overlay showing:

• Content — full text in a scrollable panel
• Metadata — ID, Type, Room, Entity, Created At, Unique flag
• Embedding preview — all dimension values to 6 decimal places
• Raw Record — expandable JSON dump of the complete database row