Building the Graph Intelligence
for Multimodal AI Agents
OpenGraph AI turns tables, text, images, audio and video into a single queryable knowledge graph — so your agent retrieves context, not chunks.
Context graph: six concepts around a core. The active concept rotates automatically; use Tab to reach a concept and Enter or Space to select it. Arrow keys also move between concepts.
Agents — Your agent calls the graph as a tool.
A memory layer your AI needs.
Benefit targets
Up to 10x faster
Aspirational target — not a proven benchmark.
Up to 40% fewer tokens
Aspirational target — not a proven benchmark.
Multimodal by design
Text, tables, images, audio, and video into one queryable context graph.
A knowledge graph
AI agent can actually query.
RAG retrieves passages. Graphs retrieve relationships. OpenGraph AI extracts entities, concepts and citations from every modality, links them, and exposes a graph your agent can traverse and reason.
Any data in
PDF, CSV, JSON, MP3/MP4, JPG/PNG, URLs, buckets, etc.. One unified pipeline.
Entities & relations
Cross-modal entity resolutions: one graph, with many senses.
Reason across
Natural language questions, graph queries, or hand it to an agent (Claude, etc.) as a tool.
The playground,
without the wait.
A live preview of what you'll build: heterogeneous sources extracted into entities, concepts and citations — explorable, queryable, agent-ready.
this is a preview — open the playground to run it
Use it where
you already work.
MCP server in Claude, CLI in terminal, REST API, or as a skill.md inside your agent runtime. Same graph — every interface.
See the full pipelineMCP Server
Connect directly inside Claude Code, Cursor, or any MCP-compatible IDE. Your editor becomes graph-aware.
# Add to Claude Code
claude mcp add opengraph -- npx -y @opengraph/mcp
# Now talk to your graph — from any IDE
extract_graph(path="./data")
query_graph("find all investors near Acme")
summarize_graph()
sync_graph_to_vector_db(target="pinecone")Your graph, where you work.
Connect an agent, install locally, or ask your graph a question.
Agent via MCP
Connected tool
Connected tool
Agent via MCP
Give Claude and other AI agents direct access to your OpenGraph knowledge.
Claude
OpenGraph MCP connected
Find the relationship between the park photos and prepare context for my agent.
I found 5 entities and 5 relationships. The lake connects the path, bench, tree, and park across all three sources.
Vector RAG retrieves nearest-neighbor passages. OpenGraph AI extracts entities, concepts and relations from every modality and lets your agent traverse them — so multi-hop questions return ranked subgraphs with a citation path, not a wall of chunks.