Repolode
Submit
bibinprathap

bibinprathap/VeritasGraph

Open-source GraphRAG and agent framework that answers multi-hop questions over a knowledge graph with citations, running locally via Ollama or in the cloud.

View on GitHubWebsite
Stars
326
Forks
39
Open issues
2
Last commit
2 days ago

What is it?

VeritasGraph builds a knowledge graph from your documents and answers questions by traversing relationships between entities, rather than only matching text by similarity. Each answer is tied to [doc#chunk] source citations, and the project says it abstains with 'insufficient evidence' when the graph lacks support.

The project ships a pip-installable Python package with a CLI, a Python API for PDF ingestion and querying, and an MCP server that exposes ingest and query tools to IDE agents such as Claude Desktop and Cursor. A local Studio web workspace (FastAPI) lets you build graphs, create agents with tools, memory, guardrails, and PII redaction, and inspect each pipeline stage. Models can run through Ollama or any OpenAI-compatible provider.

The repository also includes domain examples: a clinical knowledge graph with de-identification and ICD-10/RxNorm/SNOMED/LOINC normalization, a municipality incident chatbot, and VeritasReason, a rule-based engine for policy compliance checks.

Key features

  • 01Multi-hop question answering over a knowledge graph with [doc#chunk] citations.
  • 02Hierarchical tree navigation of document structure combined with graph search.
  • 03MCP server exposing ingest and query tools to IDE agents.
  • 04Local Studio workspace for building graphs and governed agents with guardrails and memory.
  • 05Works with Ollama and any OpenAI-compatible model provider.
  • 06Interactive PyVis graph visualization of entities and relationships.

Use cases

  • Answering questions over internal documents where the answer depends on how facts connect.
  • Running a document Q&A or agent setup fully on-premise with a local Ollama model.
  • Connecting an IDE assistant such as Cursor or Claude Desktop to a private knowledge graph.
  • Prototyping compliance or policy-check workflows over structured records.

Quick start

  1. 1

    Install the package from PyPI

    pip install veritasgraph
  2. 2

    Run the interactive demo in lite mode (needs a cloud API key, such as OPENAI_API_KEY)

    veritasgraph demo --mode=lite
  3. 3

    Run fully offline with Ollama (needs Ollama and about 8GB RAM)

    veritasgraph demo --mode=local --model=llama3.2

Things to know

  • The quick demo needs a paid OpenAI or Anthropic API key.
  • Offline mode needs a local Ollama install and roughly 8GB of RAM.
  • The full pipeline requires Docker and a Neo4j database.

In collections

Private AI knowledge bases and RAGOpen-source AI agent frameworks & appsMCP serversRun LLMs locally

Topics

#graphrag#knowledge-graph#rag#llm#multi-hop-reasoning#mcp#ollama#on-premise#agent#ai-agents

FAQ

What is VeritasGraph?
Open-source GraphRAG and agent framework that answers multi-hop questions over a knowledge graph with citations, running locally via Ollama or in the cloud. VeritasGraph builds a knowledge graph from your documents and answers questions by traversing relationships between entities, rather than only matching text by similarity. Each answer is tied to [doc#chunk] source citations, and the project says it abstains with 'insufficient evidence' when the graph lacks support.
Is VeritasGraph free to use? What license does it have?
VeritasGraph is open source under the MIT license. Check the license terms before using it in commercial products.
Is VeritasGraph still maintained?
Actively maintained. The last commit to VeritasGraph was on Nov 6, 2026.
What can I use VeritasGraph for?
Answering questions over internal documents where the answer depends on how facts connect. Running a document Q&A or agent setup fully on-premise with a local Ollama model. Connecting an IDE assistant such as Cursor or Claude Desktop to a private knowledge graph. Prototyping compliance or policy-check workflows over structured records.
How do I install VeritasGraph?
Install the package from PyPI: pip install veritasgraph Run the interactive demo in lite mode (needs a cloud API key, such as OPENAI_API_KEY): veritasgraph demo --mode=lite Run fully offline with Ollama (needs Ollama and about 8GB RAM): veritasgraph demo --mode=local --model=llama3.2
Where can I get VeritasGraph?
The source code and installation instructions are on GitHub: https://github.com/bibinprathap/VeritasGraph

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects