What is a context graph?

A context graph is a living record of decision traces stitched across entities and time, so precedent becomes searchable. Unlike traditional databases that store what happened, context graphs capture why decisions were made — the exceptions, overrides, precedents, and cross-system context that currently live in Slack threads, deal desk conversations, and people's heads. Foundation Capital calls this "AI's trillion-dollar opportunity."

What is the difference between a context graph and a knowledge graph?

Knowledge graphs store static relationships between entities (facts), while context graphs capture dynamic decision traces including the reasoning, exceptions, and temporal context behind each decision. Context graphs are optimized for AI agent decision-making and include provenance tracking, confidence levels, and reasoning traces. A knowledge graph tells you "Customer X has a gold tier account." A context graph tells you "Customer X was upgraded to gold tier because VP of Sales approved an exception based on the Q3 incident resolution precedent."

Why do AI agents need context graphs?

Without context graphs, an AI agent is like an extremely smart intern on day one — it can follow written rules but gets tripped up by every unwritten exception. Context graphs provide the institutional memory and decision precedent that AI agents need to handle edge cases and make autonomous decisions. As enterprises shift from "systems of record" to "systems of agents," the context graph becomes the crown jewel — the real source of truth for AI autonomy.

What is context engineering?

Context engineering is the natural evolution from prompt engineering. While prompt engineering focuses on crafting the right instructions for an LLM, context engineering is about curating and maintaining the optimal set of tokens (information) during LLM inference. As Anthropic defines it: "Prompt engineering adjusts instructions; context engineering controls the evidence that the instructions operate on." Context engineering is emerging as the #1 skill for AI builders in 2026.

What is a decision trace?

A decision trace is a record that captures not just what decision was made, but the specific context, reasoning, exceptions, approvals, and precedents that led to that outcome. For example, instead of recording "customer got a 25% discount," a decision trace captures "customer got a 25% discount because Service Impact Exception was invoked, approved by VP Finance, based on precedent from Q4, due to critical incidents." Decision traces are the core data structure within a context graph.

Open Source Context Graph Tools: A Comprehensive Guide (2026)

Q: What are the best open source context graph tools?

The leading open source context graph tools in 2026 include: TrustGraph (Apache 2.0, ontology-driven graph construction with OntologyRAG), Graphiti by Zep (45,000+ GitHub stars, bi-temporal knowledge graph with P95 latency of 300ms), and Neo4j GraphRAG (the industry standard graph database with dedicated RAG pipeline tools). Other notable tools include Graphlit (cloud-native context layer), DataHub Agent Context Kit, and the Agent Trace specification by Cognition AI and Cursor.

Introduction

Open source tools are driving the context graph movement. As organizations recognize that standard RAG pipelines fall short of delivering truly context-aware AI, a new generation of tools has emerged to fill the gap. These projects go beyond simple vector search, offering structured knowledge representation, temporal awareness, and ontology-driven extraction that traditional retrieval systems cannot match.

The landscape in 2026 is remarkably mature. Projects that were experimental just a year ago now serve production workloads. From temporal knowledge graphs that track how facts evolve over time, to ontology-driven graph construction that uses domain expertise to guide knowledge extraction, the ecosystem offers options for every architecture and use case.

What makes this moment particularly significant is the convergence around interoperability standards. The Model Context Protocol (MCP) has become the connective tissue linking these tools to AI agents, and most leading projects now offer MCP support out of the box. This means you can evaluate tools based on their core strengths rather than worrying about integration compatibility.

“The shift from monolithic RAG pipelines to composable context graph infrastructure is the most important architectural change in AI since the transformer.”

This guide provides an in-depth look at the three most impactful open source context graph tools—TrustGraph, Graphiti, and Neo4j GraphRAG—along with a survey of other notable projects, a feature comparison table, and a decision framework to help you choose the right tool for your needs.

TrustGraph

TrustGraph positions itself as the “Context Operating System for AI”—an ambitious framing that reflects its comprehensive approach to context graph construction. Released under the Apache 2.0 license, TrustGraph is fully open source and designed for production deployments across cloud environments.

OntologyRAG

TrustGraph coined the term “OntologyRAG”—a methodology that guides knowledge extraction using domain-specific ontology definitions. Rather than extracting unstructured text chunks, OntologyRAG uses a formal ontology to structure how knowledge is identified, categorized, and connected. This means your context graph reflects your domain model, not just the surface content of your documents.

Core Features

Ontology-Driven Graph Construction

Define your domain ontology and TrustGraph will use it to guide entity extraction, relationship mapping, and knowledge organization. This prescribed approach ensures that the resulting graph structure aligns with your domain expertise and organizational needs.

Custom Ontology Engineering

Build and iterate on domain-specific ontologies that capture the entities, relationships, and constraints unique to your industry. TrustGraph provides tooling for ontology creation, validation, and versioning.

Automated Context Graph Construction

Ingest documents, extract structured knowledge guided by your ontology, and automatically populate your context graph. The pipeline handles entity resolution, relationship extraction, and graph population.

Intelligent Context Retrieval

Query the graph using natural language. TrustGraph translates queries into graph traversals that respect ontological structure, returning contextually rich answers rather than raw text chunks.

MCP Interoperability

Native support for the Model Context Protocol enables AI agents to interact with TrustGraph as a context source, making it straightforward to integrate with agent frameworks and orchestration tools.

3D Visualization

Explore your context graph visually in three dimensions. This aids in ontology debugging, knowledge gap identification, and stakeholder communication.

Context Cores

One of TrustGraph's distinguishing concepts is Context Cores—packaged and versioned units of processed context. A Context Core bundles the ontology, extracted knowledge, and graph structure into a portable artifact that can be shared, versioned, and deployed independently. This makes it possible to treat context as a first-class deliverable, much like container images for application code.

Integrations

LLM Providers

• OpenAI
• Anthropic
• AWS Bedrock
• Google VertexAI
• Private LLM inference

Graph & Vector Stores

• Neo4j
• Qdrant
• Pinecone

Getting Started

TrustGraph provides multiple entry points for new users. The documentation at docs.trustgraph.ai covers installation, ontology engineering, and deployment. The Configuration Builder offers a guided setup experience, and the Workbench (available on port 8888) provides an interactive environment for ontology development, graph exploration, and query testing. Multi-cloud deployment is supported out of the box, including air-gapped environments with private LLM inference.

GitHub:

trustgraph-ai/trustgraph

Graphiti (by Zep)

Graphiti is a framework for building temporally-aware knowledge graphs, developed by the team at Zep. With over 45,000 GitHub stars, it has become one of the most widely adopted open source projects in the context graph space. Graphiti's core innovation is making time a first-class citizen in knowledge graph construction and querying.

Bi-Temporal Data Model

Graphiti implements a bi-temporal data model that tracks two distinct time dimensions: event time (T), when something actually happened in the real world, and ingestion time (T'), when the system learned about it. This distinction is critical for accurate reasoning about temporal context—knowing not just what is true, but when it became known.

Key Capabilities

Real-Time Incremental Updates

Unlike batch-oriented systems that require full recomputation when data changes, Graphiti handles incremental updates in real time. New information is integrated into the graph without disrupting existing structure, making it ideal for applications where data arrives continuously.

Hybrid Search

Graphiti combines three search modalities into a unified retrieval system: semantic embeddings for meaning-based search, keyword search for precise term matching, and graph traversal for relationship-aware context retrieval. This hybrid approach consistently outperforms single-modality systems.

MCP Server Implementation

Graphiti provides a full MCP server implementation, allowing AI agents to query temporal knowledge graphs through the standard Model Context Protocol interface. This makes Graphiti a natural fit for agent memory architectures.

Agent Memory

Graphiti is purpose-built for AI agent memory use cases. It enables agents to maintain long-term context across conversations, remember previous interactions, and reason over accumulated knowledge with temporal awareness.

Performance Benchmarks

Graphiti has published compelling benchmark results that demonstrate its effectiveness:

94.8%

DMR benchmark accuracy (vs 93.4% MemGPT)

300ms

P95 retrieval latency

18.5%

Accuracy improvement on LongMemEval

Supported Infrastructure

Graphiti supports Neo4j and FalkorDB as its graph database backends. The project has published a research paper on arXiv (2501.13956) detailing the bi-temporal architecture and benchmark methodology.

GitHub:

getzep/graphiti

Neo4j GraphRAG

Neo4j has long been the graph database standard for knowledge graphs, and their GraphRAG Python Package extends that foundation into the world of AI-powered retrieval and generation. For teams already invested in the Neo4j ecosystem, this is the natural path to building context graph infrastructure.

The GraphRAG Pipeline

Neo4j's GraphRAG package provides a structured pipeline for building GenAI applications on top of knowledge graphs:

1Entity Extraction

Identify entities and their types from unstructured text using LLMs. The extraction process is configurable, allowing you to define custom entity types and extraction rules that match your domain.

2Embedding Generation

Generate vector embeddings for entities and their properties, enabling semantic similarity search across the knowledge graph. Neo4j's native vector index support makes this efficient at scale.

3Knowledge Graph Creation

Populate the Neo4j graph with extracted entities, relationships, and embeddings. The resulting knowledge graph serves as the foundation for context-aware retrieval and reasoning.

Key Strengths

●Multi-Hop Reasoning: Unlike flat vector retrieval, Neo4j GraphRAG can traverse multiple relationships to find relevant context. “Find all customers affected by the supplier issue” requires traversing customer → product → component → supplier relationships—something vector search alone cannot do.
●Explainability: Graph-based retrieval provides natural explainability. You can trace exactly which relationships led to a particular answer, making it easier to build AI audit trails and satisfy compliance requirements.
●Scalability: Neo4j handles billions of nodes and relationships in production, with enterprise clustering, ACID transactions, and proven operational tooling.
●MCP Server: Neo4j offers an MCP Server for AI agent integration, allowing agents to query and traverse the knowledge graph through the standard protocol.

Learning Resources

Neo4j provides extensive free educational resources through GraphAcademy, including courses on knowledge graph construction, Cypher query language, graph data science, and GenAI application development. For teams evaluating Neo4j GraphRAG, these courses offer a structured path from graph fundamentals to production deployment.

Package:

neo4j GraphRAG Python Package

Other Notable Tools

Beyond the three major platforms, several other projects deserve attention for their unique approaches to context graph infrastructure:

Graphlit

A cloud-native context layer that provides one API for the full stack of context graph operations. Graphlit handles ingestion, extraction, graph construction, and retrieval through a single unified API, reducing the integration burden for teams that want context graph capabilities without assembling multiple components. Its cloud-native architecture means zero infrastructure management.

DataHub Agent Context Kit

Built on the popular DataHub metadata platform, the Agent Context Kit provides MCP tools for metadata interaction. This is particularly valuable for data teams that already use DataHub for data cataloging and want to make their metadata graph accessible to AI agents. It bridges the gap between data governance infrastructure and AI agent workflows.

ApeRAG

A production-ready GraphRAG implementation that focuses on reliability and ease of deployment. ApeRAG provides a streamlined path from documents to graph-augmented retrieval, with emphasis on operational stability and straightforward configuration for teams that need GraphRAG capabilities without extensive customization.

Semantica

A platform focused on the semantic layer and knowledge engineering aspects of context graphs. Semantica provides tooling for building and managing ontologies, defining semantic relationships, and ensuring consistency across knowledge domains. It is well-suited for organizations with complex domain models that require formal semantic rigor.

PowerMem

An AI-powered long-term memory system designed to give AI agents persistent, structured memory across sessions. PowerMem focuses on the agent memory problem specifically, providing efficient storage, retrieval, and reasoning over accumulated conversational and operational context.

Feature Comparison

The following table compares the three major open source context graph tools across key dimensions:

Feature	TrustGraph	Graphiti	Neo4j GraphRAG
License	Apache 2.0	Apache 2.0	Apache 2.0
Primary Use Case	Ontology-driven context graphs	Temporal agent memory	Graph-augmented RAG
MCP Support	Yes	Yes (MCP server)	Yes (MCP server)
Temporal Awareness	Basic	Bi-temporal (T and T')	Via application layer
Ontology Support	Native (OntologyRAG)	Schema-flexible	User-defined labels
Graph Database	Neo4j, Qdrant, Pinecone	Neo4j, FalkorDB	Neo4j
Search Approach	Ontology-guided retrieval	Hybrid (semantic + keyword + graph)	Graph traversal + vector
LLM Providers	OpenAI, Anthropic, Bedrock, VertexAI, Private	OpenAI, Anthropic, others	OpenAI, Anthropic, others
Visualization	3D graph visualization	Via Neo4j Browser/Bloom	Neo4j Browser, Bloom
Deployment	Multi-cloud, air-gapped	Self-hosted, cloud	Self-hosted, Aura cloud

All three projects are actively maintained and have strong community support. The choice between them depends less on absolute capability and more on which approach best fits your specific architecture, team expertise, and use case requirements.

How to Choose

Selecting the right context graph tool depends on your specific requirements, existing infrastructure, and team capabilities. Use this decision framework to narrow your options:

Choose TrustGraph if…

✓You have well-defined domain ontologies or need to build custom ones
✓OntologyRAG's structured extraction approach matches your data
✓You need multi-cloud or air-gapped deployment with private LLM inference
✓Context Cores packaging and versioning aligns with your deployment model
✓You want 3D visualization for ontology exploration and debugging

Choose Graphiti if…

✓Temporal awareness is a core requirement (bi-temporal data model)
✓You need real-time incremental updates without batch recomputation
✓Agent memory is your primary use case (long-term context for AI agents)
✓Low-latency retrieval matters (P95 of 300ms)
✓You want hybrid search combining semantic, keyword, and graph traversal

Choose Neo4j GraphRAG if…

✓You already use Neo4j or have graph database expertise on your team
✓Multi-hop reasoning across complex relationship chains is critical
✓Explainability and audit trails are regulatory requirements
✓You need enterprise-grade scalability (billions of nodes)
✓Free learning resources (GraphAcademy) are important for team onboarding

A Pragmatic Note

These tools are not mutually exclusive. Many production architectures combine them—for example, using Neo4j as the graph database backend while leveraging Graphiti's temporal layer for agent memory, or using TrustGraph's ontology engineering for knowledge extraction and Neo4j for storage and querying. The MCP standard makes this kind of composition increasingly practical.

Regardless of which tool you choose, the most important step is to start building. The context graph ecosystem is evolving rapidly, and teams that begin accumulating structured knowledge now will have a compounding advantage over those that wait. Pick the tool that best fits your current needs and iterate from there.

Frequently Asked Questions

What are the best open source context graph tools in 2026?

The leading open source context graph tools in 2026 include TrustGraph (the Context Operating System with OntologyRAG), Graphiti by Zep (temporally-aware knowledge graphs with 45k+ GitHub stars), and Neo4j GraphRAG (the graph database standard for knowledge graph-powered RAG). Other notable tools include Graphlit, ApeRAG, Semantica, DataHub Agent Context Kit, and PowerMem.

What is OntologyRAG and how does it differ from standard RAG?

OntologyRAG is a methodology coined by TrustGraph that guides knowledge extraction using domain-specific ontology definitions. Unlike standard RAG, which retrieves raw text chunks based on semantic similarity, OntologyRAG uses a formal ontology to structure how knowledge is identified, categorized, and connected. This produces more accurate and contextually relevant responses because the retrieval is grounded in domain structure rather than surface-level text similarity.

Does Graphiti support temporal knowledge graphs?

Yes, Graphiti by Zep is specifically designed for building temporally-aware knowledge graphs. It uses a bi-temporal data model tracking both event time (T)—when something happened in the real world—and ingestion time (T')—when the system learned about it. This enables real-time incremental updates without requiring batch recomputation and supports sophisticated temporal queries.

Which open source context graph tools support MCP?

Multiple open source context graph tools support MCP (Model Context Protocol) in 2026. TrustGraph offers MCP interoperability for AI agent integration. Graphiti provides a full MCP server implementation for seamless agent connectivity. Neo4j also offers an MCP Server for integrating its graph database with AI agents. MCP support has become a baseline expectation for tools in this space.

How do I choose between TrustGraph, Graphiti, and Neo4j GraphRAG?

Choose TrustGraph if you need ontology-driven knowledge extraction with custom domain modeling and multi-cloud deployment. Choose Graphiti if your priority is temporal awareness, real-time incremental updates, and agent memory with low latency. Choose Neo4j GraphRAG if you need enterprise-grade graph database infrastructure with multi-hop reasoning and already use or prefer the Neo4j ecosystem. These tools can also be combined in a single architecture.

Continue Learning

IMPLEMENTATION