NeuroTIC: An Autonomous Knowledge Graph Construction Engine Based on Dual-Agent Architecture

NeuroTIC is an open-source multi-agent ontological engine. Through the collaboration of exploration agents and expansion agents, combined with the CozoDB graph database and Ollama local large language model, it achieves the automated construction of complex knowledge networks from a single concept seed. The project aims to solve the problems of high cost and heavy reliance on manual work in traditional knowledge graph construction, lowering the threshold for knowledge engineering.

In the era of information explosion, effectively organizing and understanding complex domain knowledge has become a key challenge. Traditional knowledge graph construction relies heavily on manual annotation and domain experts, which is costly and difficult to scale. NeuroTIC emerged to address this, aiming to enable machines to automatically explore, associate, and expand a complete knowledge network starting from a small number of seed concepts through an autonomous agent system.

NeuroTIC adopts a dual-agent architecture:

• Exploration Agent: Responsible for navigating the knowledge space, discovering new concepts and relationships, and identifying directions worth exploring in depth—similar to the literature review phase for researchers.
• Expansion Agent: Focuses on deepening discovered concepts, mining detailed attributes of nodes, related concepts, and context, and integrating them into the graph—corresponding to the in-depth analysis and integration phase for researchers.

NeuroTIC's tech stack selection balances privacy, cost, and scalability:

• CozoDB: An embedded graph database that supports Datalog queries. It is lightweight and high-performance, suitable for local deployment and prototype verification, with concise and elegant complex relationship queries.
• Ollama: Integrates local open-source large language models, ensuring data privacy, zero API call costs, supporting hot-swapping of models, and flexibly adapting to task complexity.

NeuroTIC's operation process is divided into stages:

1. Seed Injection: Users provide initial concepts or documents, which are parsed into seed nodes and injected into the graph.
2. Iterative Exploration: The exploration agent proposes expansion suggestions based on the current graph and identifies high-value directions.
3. Deep Expansion: The expansion agent conducts detailed analysis of new nodes, extracts attributes, establishes relationships, and generates descriptions.
4. Quality Verification: Built-in consistency checks ensure the logical self-consistency of newly added knowledge.
5. Iterative Optimization: Continuous iteration leads to the gradual enrichment and improvement of the graph.

NeuroTIC's application potential covers multiple domains:

• Academic Research: Quickly build a knowledge network of literature on research topics, discovering gaps and cross-disciplinary opportunities.
• Enterprise Knowledge Management: Automatically extract organizational knowledge from scattered internal documents and build an exclusive asset library.
• Personal Learning: Input domain keywords to generate structured learning paths and concept maps.

As a fully open-source project, NeuroTIC not only provides runnable code but also demonstrates a new knowledge engineering paradigm. It proves that through agent collaboration and local AI capabilities, the threshold for knowledge graph construction can be significantly lowered, providing an excellent starting point and reference implementation for developers and researchers.