We just launched VeritasGraph, an open-source RAG framework that swaps out pure vector search for a knowledge graph backbone—and the response has been exciting: 3,300+ visitors and 130 GitHub stars in just 5 days.
Why we built it:
Vector RAG ≠ enough → Great for simple lookups, but it breaks on multi-hop questions that need connecting facts across docs.
Verifiable attribution → Every AI output is traced to its original source, making results auditable and reducing hallucinations.
Data sovereignty → 100% on-prem, no vendor lock-in, no third-party data exposure.
We want enterprise AI to be trustworthy, private, and explainable—not just “accurate enough.”
Repo: github.com/bibinprathap/VeritasGraph
Would love HN’s thoughts:
Are you running into the limits of vector-only RAG?
How are you tackling explainability and trust in your AI apps?
Bibinprathap•1h ago
Our immediate roadmap is focused on:
Expanded LLM Support: We're working on adding support for more open-source models (like the latest from Mistral and Llama) as well as improving integration with closed-source APIs for teams that use them in a hybrid environment.
More Data Connectors: We're prioritizing connectors for common enterprise knowledge sources like Confluence, SharePoint, and Salesforce to make ingestion seamless.
Performance Benchmarking: We're building a public benchmark to quantitatively show the performance lift of Graph RAG on complex, multi-hop questions compared to traditional vector-search RAG.
Simplified Deployment: We're finalizing a one-click Docker Compose setup and a Helm chart for easier deployment on Kubernetes.
We're building this in the open and would love to hear what the community thinks we should prioritize. What features would make this most useful for you?