Simple, Transparent Pricing

What counts as a "node"?

A node is a single vector embedding in your database — typically corresponding to one text chunk. If your vector database has 100,000 embeddings, that's 100,000 nodes. Metadata and edges don't count toward your node limit.

Does refinement delete my original data?

No. NodeRefine operates on a shadow copy by default. Pruned nodes are archived (not deleted) and can be restored at any time. In Enterprise plans, you can enable "write-through" mode which modifies the source database directly, but even then, a full snapshot is taken before each refinement job.

Is NodeRefine HIPAA compliant?

Yes. Our Enterprise plan includes HIPAA, SOC 2 Type II, and GDPR compliance modes. The Medical template automatically detects and protects PHI-tagged nodes, and every refinement action is logged in an immutable audit trail. We sign BAAs for qualifying customers.

Can I use NodeRefine with my existing RAG framework?

Absolutely. We publish official plugins for LangChain and LlamaIndex. For other frameworks, our REST API and Python/Rust SDKs integrate with any pipeline. NodeRefine works as a middleware layer — it sits between your vector database and your application.

How long does a refinement job take?

It depends on your node count and strategies. A full three-stage refinement of 1M nodes typically completes in under 5 minutes. The de-duplication stage is the most compute-intensive due to cross-encoder scoring. You can track progress in real-time via the status API or webhooks.

What's the difference between "real-time sync" and "write-path" refinement?

Real-time sync (Pro) monitors your vector database and triggers refinement jobs whenever significant changes are detected — typically within minutes. Write-path refinement (Enterprise) intercepts every write operation to your vector database and refines the data inline before it's committed — achieving sub-second optimization at the point of ingestion.