​Context graphs as the control plane for the agentic enterprise, Dave Bennett, Indykite

Indykite proposes using context graphs as a dynamic control plane to solve the fragmentation and security risks of multi-agent systems in the enterprise. By decoupling agent intelligence from governance, organizations can update their AI models without rebuilding their entire policy framework. Unlike traditional knowledge graphs that only store static facts, context graphs incorporate provenance, temporal validity, and decision tracing. This allows every action an agent takes to be materialized in the graph, making it possible to query exactly why and how a specific decision was made.

The strategy shifts governance from static, role-based labels to Knowledge-Based Access Control (KBAC), which evaluates intent, data sensitivity, and trust signals in real-time. By giving agents formal identities within the graph, Indicite aims to move the enterprise from merely governing data to operationalizing it safely. This centralized approach ensures that as companies scale from a few agents to thousands, they maintain human oversight, identity verification, and fine-grained access control across all internal and external tools.

In this talk from the Context Graph Meetup, Indykite’s Dave Bennett shows how to:
• Decouple agent intelligence from agent governance so your AI stack can evolve without rebuilding policies
• Give AI agents formal identities in the context graph, with dynamic Knowledge-Based Access Control (KBAC)
• Implement full delegation chains using OAuth token exchange — so every agent action traces back to a human
• Move from static role-based permissions to real-time, intent-aware authorization for multi-agent workflows

0:00 Introduction: why enterprises need a context graph control plane for AI
2:00 Decoupling agent intelligence from governance — why it matters at scale
5:00 How context graphs go beyond knowledge graphs: provenance and temporal validity
8:00 Reference architecture: agent gateway, orchestrator, and retriever with Indykite
13:00 Live demo: a CSR agent uses the context graph to make a grounded customer decision
18:00 Knowledge-Based Access Control (KBAC): scope down agent permissions in real time
21:00 How agent context changes what data is accessible — even for the same organization

Get started with context graphs in Neo4j: https://graphacademy.neo4j.com/