From documents to a living ontology
Spyndel works like a spider building a web: every new strand is anchored to what is already there. Here is what happens to your documents, step by step.
Read
Documents are broken into individual statements, such as requirements, specifications and facts. Each keeps where it came from: document, page and section heading.
Understand
Each statement is mapped into a meaning space, so statements about the same subject sit together even when they use different words. Boilerplate, citations and headers are recognized and set aside.
Organize
Groups of related statements become topics, and topics roll up into categories, giving you a browsable ontology named in plain language. It is discovered from your content, not imposed on it.
Connect
Candidate connections between statements are checked one by one. Statements that only look alike, such as the same template about a different component, are rejected. Each connection that passes gets a calibrated confidence score.
Serve
The graph lives in a purpose-built store designed for fast lookups and traversal. People browse and ask; AI assistants connect through MCP; analysts query it with Silk.
Learn
When questions are answered, Spyndel records which statements were cited and which connections led to them. Useful paths rise; noise fades.
Small specialist models, not a token meter
Most AI knowledge-graph tools send every page to a large language model to extract entities and relationships, and charge you per token, forever. Spyndel's engine uses compact models trained for one job each: recognizing substance, understanding meaning and verifying connections.
- No per-document LLM calls when documents are added and linked.
- Runs on hardware you already have: CPU anywhere, and much faster on a GPU.
- Predictable cost: a bigger library means more compute time, not a bigger AI bill.
- AI where it adds value: optional document summaries and question answering use the provider you choose.
Every answer shows its work
An answer is only as good as its sources. Spyndel checks each citation before you see it: the statement exists, it was actually retrieved, and the quoted words appear in it. Unverified citations are flagged, not hidden.
For each citation you get
- The exact statement and the text around it
- Document, page and section
- The topic and category it belongs to
- A link to open the source at the cited page