Why your agents need decision traces, not just documents — Zach Blumenfeld, Neo4j

2026-05-29 · Source: AI Engineer · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, long

Summary

Neo4j's Zach Blumenfeld introduces context graphs, a knowledge layer graph database concept designed to enhance AI agent accuracy and explainability by providing decision traces. Unlike traditional knowledge bases for RAG, context graphs integrate past decisions, precedents, and dynamic information to enable agents to make better decisions, as demonstrated with a financial analyst agent example. The system uses Claude for agent runtime, OpenAI embeddings, Neo4j for vector assistance, and a Next.js frontend. Key tools include the "create context graph" command-line utility, which scaffolds full-stack applications with pre-built domains (e.g., healthcare, FinServ) or custom ontologies, and the open-source Neo4j agent memory package, offering a complete memory API with short-term, long-term, and reasoning capabilities, including a text-to-knowledge graph process using spaCy, gliner, and LLM fallback.

Key takeaway

For AI Engineers building intelligent agents that require explainable decision-making, integrating context graphs is crucial. Your systems can move beyond simple question-answering to provide justified accept/reject decisions by incorporating past decision traces and causal chains. Explore Neo4j's "create context graph" tool to quickly scaffold agent applications and leverage the "agent memory" package for robust short-term, long-term, and reasoning capabilities, enhancing agent autonomy and reliability.

Key insights

Context graphs provide decision traces and causal chains, enabling AI agents to make better, explainable decisions.

Principles

• Integrate past decisions and precedents for agent reasoning.
• Combine semantic and structural similarity for hybrid search.
• Embed graph nodes into vectors for similar decision trace lookup.

Method

The Neo4j agent memory package converts raw text to a knowledge graph via spaCy, gliner, LLM fallback, followed by merging, deduplication, and enrichment.

In practice

• Use "create context graph" for rapid full-stack agent app development.
• Import data from GitHub, Notion, Jira, or Slack into context graphs.
• Generate custom graph ontologies for specific domains.

Topics

• Context Graphs
• Neo4j
• AI Agents
• Knowledge Graphs
• Graph Embeddings
• Decision Traces
• LLM Memory

Best for: NLP Engineer, AI Engineer, Machine Learning Engineer, AI Architect

Related on AIssential

• Context Graphs for Explainable, Decision-Aware AI Agents — Andreas Kollegger & Zaid Zaim, Neo4j — Context graphs empower AI agents with decision-making capabilities by integrating knowledge, rules, and policies.
• Context Graphs: AI's Next Big Idea — Context graphs capture the "why" behind enterprise decisions, enabling AI agents to understand and apply organizational precedents.
• Connecting the Dots with Context Graphs — Stephen Chin, Neo4j — Context graphs integrate LLMs with knowledge graphs to provide grounded, explainable, and auditable AI-driven decision-making by connecting disparate enterprise data.
• Context Graphs: AI's Next Big Idea — Context graphs capture decision traces, the "why" behind actions, essential for scalable AI agent autonomy in enterprises.
• Context Graphs: A Series of Unfortunate Events — Context graphs model information as a series of time-bound events, contrasting with topical knowledge graphs.
• LLM Friendly Graphs for AI Memory — From Triples to Facts — Larger, context-rich labels and subgraph representations enhance LLM comprehension of complex facts in knowledge graphs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Engineer.