Veritasgraph
VeritasGraph — open-source Knowledge Graph & GraphRAG framework on GitHub. Build multi-hop reasoning, ontology-aware retrieval, and verifiable attribution over your own data. Nodes, edges, RDF, linked-data — runs locally or in the cloud.
Last verified:
What is Veritasgraph?
VeritasGraph is an enterprise-grade, open-source Graph RAG (Retrieval-Augmented Generation) framework that transforms unstructured content—such as YouTube videos, web articles, PDFs, and documents—into structured, queryable Knowledge Graphs. Unlike traditional vector-search RAG systems that suffer from context blindness, VeritasGraph uses a knowledge graph backbone to enable multi-hop reasoning, revealing hidden relationships between entities across documents.
Key features include instant URL ingest for auto-extracting transcripts and entities, verifiable attribution that grounds every answer with source citations to eliminate hallucinations, global and local search modes for community-wide reasoning or entity-specific details, an interactive Graph Explorer for real-time data visualization, and precision control with adjustable temperature and preset output styles (Bullet, Detailed, Focused). The framework includes demo datasets for student visa eligibility, athlete recruitment analytics, and project management dashboard analysis.
VeritasGraph is designed for enterprise AI teams, data scientists, developers building RAG pipelines, CTOs and CISOs prioritizing data sovereignty, researchers needing trustworthy AI with auditable evidence chains, and organizations deploying on-premise AI without vendor lock-in. It runs entirely on local hardware using Ollama with Llama 3.1 or other local LLMs, ensuring complete data privacy and control.
The framework addresses critical barriers to enterprise AI adoption: vector RAG不足以 handle multi-hop questions, unverified outputs causing hallucinations, and third-party data exposure risks. By providing explainable AI with full source traceability, VeritasGraph enables trustworthy, secure, and transparent AI systems for scientific knowledge discovery, literature review, complex data analysis, and business intelligence.
Built with a FastAPI backend, Next.js frontend with enterprise authentication, Gradio UI for easy setup, and Model Context Protocol (MCP) server for secure data actions, VeritasGraph integrates with Power BI for natural language queries and automatic DAX generation, making it suitable for AI-first analytics experiences.
Veritasgraph pricing
Pricing model: Freemium
VeritasGraph is completely free and open-source under the MIT license. The open-source plan includes unlimited local use, the full framework/library, self-hosted deployment, community support, and all core features. There are no setup fees, no subscription costs, and no paid tiers. Users can run it entirely on their own hardware using local LLMs like Llama 3.1 via Ollama. Token costs depend on the user's chosen LLM provider—the indexing process makes many LLM calls, so for a 1M token corpus expect roughly 5-10x the token cost of the source material. Users can reduce costs by using local inference servers (Ollama, vLLM) instead of cloud APIs.
Veritasgraph pros
- Open-source with MIT license for free forever local use
- 100% on-premise deployment ensures complete data sovereignty
- Verifiable attribution links every answer to original source text
- Eliminates hallucinations through grounded, traceable outputs
- Multi-hop reasoning connects facts across multiple documents
- Instant URL ingest auto-extracts YouTube transcripts and web entities
- Interactive Graph Explorer visualizes data connections in real-time
- Supports local LLMs like Llama 3.1 via Ollama for cost reduction
- Global and local search modes for flexible reasoning approaches
- Adjustable temperature and preset output styles for precision control
- No vendor lock-in or third-party data exposure risks
- Includes demo datasets for immediate testing and validation
- FastAPI backend with Next.js frontend and enterprise authentication
- Model Context Protocol server enables secure structured data actions
- Power BI integration for natural language queries and DAX generation
- Incremental indexing allows adding documents without full reprocess
- Gradio UI provides easy setup for non-technical users
Veritasgraph cons
- Requires local hardware with sufficient resources for running LLMs
- Indexing makes many LLM calls resulting in 5-10x token costs
- Full re-indexing may be needed for optimal community detection
- Setup complexity for non-technical users despite Gradio UI
- Depends on Ollama or compatible local inference servers
- Limited documentation compared to commercial RAG solutions
- No cloud-hosted managed service option available
- Community support only without enterprise SLA guarantees
Frequently asked questions about Veritasgraph
What is VeritasGraph and how does it differ from traditional RAG?
VeritasGraph is an enterprise-grade Graph RAG framework that swaps pure vector search for a knowledge graph backbone. Traditional RAG retrieves relevant text chunks via vector similarity alone, but VeritasGraph first builds a knowledge graph capturing entities and relationships from your documents, then uses graph structure plus community summaries for retrieval. This enables multi-hop reasoning and global sensemaking that traditional vector-search RAG cannot handle, especially for complex questions requiring connecting facts across multiple documents.
How does verifiable attribution work in VeritasGraph?
VeritasGraph is designed to trace every part of the generated answer back to the original source text, providing verifiable attribution. When the system generates a response, each assertion is linked back to the specific sections of source documents that contributed to it. This creates a clear, auditable trail that helps mitigate hallucinations and allows users to verify the accuracy and origin of AI-generated insights, making results trustworthy and transparent.
Can I use VeritasGraph with my own local LLMs?
Yes, VeritasGraph supports any OpenAI-compatible API endpoint, so you can use Ollama, vLLM, or other local inference servers. It runs entirely on your own hardware using local LLMs like Llama 3.1 via Ollama, ensuring complete data privacy. This allows you to reduce costs significantly compared to using cloud-based API providers while maintaining full control over your data and AI models.
What types of content can VeritasGraph ingest?
VeritasGraph can ingest unstructured content including YouTube videos (auto-extracts transcripts), web articles (via URL paste), PDFs, text documents, and other document formats. The instant ingest feature allows you to paste a URL for YouTube or web content to automatically extract transcripts and entities. The framework builds a knowledge graph from these documents by extracting key entities and their relationships using an LLM during the indexing process.
How does the indexing process work and what are the costs?
During indexing, VeritasGraph ingests your documents and uses an LLM to automatically build a knowledge graph by extracting key entities and their relationships. The process makes many LLM calls for entity extraction and summarization. For a 1M token corpus, expect roughly 5-10x the token cost of the source material. The tradeoff is dramatically better retrieval quality. VeritasGraph supports incremental indexing, allowing you to add new documents without reprocessing the entire corpus, though full re-indexing may be needed for optimal community detection.
What are the global and local search modes?
VeritasGraph offers two search modes you can toggle between: global search enables community-wide reasoning across the entire knowledge graph, while local search focuses on entity-specific details. Global search uses community summaries for broad sensemaking, whereas local search uses graph traversal to gather context around specific entities. This flexibility allows you to choose the appropriate reasoning approach based on whether you need broad overview insights or detailed information about particular entities.
Is VeritasGraph suitable for enterprise deployment?
Yes, VeritasGraph is specifically designed as an enterprise-grade framework. It provides 100% on-premise deployment with no vendor lock-in and no third-party data exposure, addressing critical enterprise concerns about data sovereignty and security. The framework includes enterprise-level authentication, a FastAPI backend, Next.js frontend, and Model Context Protocol server for secure structured data actions. It's built for organizations that need trustworthy, private, and explainable AI—not just accurate enough results.
What demo datasets are included with VeritasGraph?
VeritasGraph includes three demo datasets for immediate testing: Student Visa & Admission Eligibility data, Elite Athlete Recruitment Analytics data, and Project Management Dashboard Analysis data. These datasets allow you to quickly try the live demo and understand how the framework handles different types of content and queries. The live demo is available at the stable URL that always redirects to the current server.
How does VeritasGraph handle multi-hop reasoning questions?
When you ask a question, VeritasGraph uses a hybrid retrieval approach. First, vector search finds the most relevant starting points in the knowledge graph. Then it traverses the graph's relationships to gather the full context needed to answer multi-hop questions. The rich, graph-based context is fed to a LoRA-tuned LLM to generate an answer. This graph traversal enables connecting facts across multiple documents, which traditional vector-search RAG breaks on but VeritasGraph handles effectively.
How do I get started with VeritasGraph?
The complete project is available on GitHub at github.com/bibinprathap/VeritasGraph. You can explore the repository, star the project, and build your own. The GitHub repository includes a comprehensive guide for setting everything up from installing dependencies to running the indexing process. It comes with a Gradio UI for easy setup and an interactive chat experience. You can also try the live demo at bibinprathap.github.io/VeritasGraph/demo/ before setting up your own instance.