opengraph-image turns a folder of images into a knowledge graph that an agent can inspect and query conversationally.
What the package extracts
The image pipeline looks beyond filename metadata and captions. It identifies:
- Objects
- Scenes
- Attributes
- Relationships between visible entities
- The source image supporting each extraction
Package interfaces
| Interface | Best for |
|---|---|
| CLI | Local exploration, scripts, and repeatable terminal workflows |
| MCP server | Claude Code, Claude Desktop, Cursor, and other MCP-compatible agents |
graph.json |
Inspecting or integrating the serialized graph directly |
Build lifecycle
folder
├─ image-01.jpg
├─ image-02.png
└─ image-03.webp
↓ build
opengraph-out/
└─ graph.json
↓ query
entities + relationships + grounded answer
A relationship-aware query
Keyword search can locate images tagged with “dog” or “water.” A graph query can ask for the relationship between them:
Which images show a dog near a body of water?
The graph can follow the extracted near relationship and return the source images supporting the match.
Current scope
The public v1 package is focused on images. The wider OpenGraph AI project is building toward a shared semantic graph layer for tables, text, audio, and video, but documentation in this section describes the image package that is available now.