Common questions
How is a CKG different from RAG?
RAG retrieves chunks of text and lets the model infer relationships. A CKG declares relationships as typed edges — the agent traverses them deterministically. No hallucination on the path, because the path is declared, not inferred. CKG F1: 0.471 vs RAG 0.123 on the KRB benchmark.
How do I activate my key?
After checkout, your key arrives by email. Set export CKG_API_KEY=your-key-here, restart your MCP client, and list_domains() shows all unlocked domains. No reinstall needed.
What is a "Sealed Appliance"?
A Docker container with 20+ domain CKGs for a single vertical (e.g., ckg-finance-mcp). It runs on your hardware — no calls home, no cloud dependency. The knowledge graphs are auditable: you can read every edge. Ideal for regulated environments where data sovereignty is required.
Can I self-host?
Pro and Developer tiers run as an MCP server on your machine — nothing is sent to Graphify.md servers at query time. The Sealed Appliance is fully air-gapped. License key validation pings our server once per 24 hours.
Is there an enterprise license?
Yes. For teams, on-prem deployments, or custom domain graphs —
email us. We build custom CKGs for domains not yet in the catalog (24-hour turnaround, starting at $499/domain).
What agents and clients does this work with?
Any MCP-compatible client: Claude Desktop, Cursor, Windsurf, VS Code with MCP extension, custom Python agents via FastMCP, LangChain, LangGraph. Also works as a standalone Python library.