How it works
Iontos runs everything — public sources and your private ones — through one pipeline. The pipeline doesn't care where a piece of content came from; the same path turns a company registry record, a web page, or one of your internal documents into clean, connected knowledge.
The pipeline
Sources → Ingest → Extract → Resolve → Graph
| Stage | What it does |
|---|---|
| Ingest | Pulls in a document or record, archives the original, and removes duplicates of things already seen. |
| Extract | Reads the content and pulls out the entities, facts, and relationships it contains. |
| Resolve | Decides which records describe the same real-world thing, reconciles facts that disagree, and runs quality checks. |
| Graph | Commits the clean, resolved knowledge to a graph you can query. |
Two ways to read a source
Not all content is the same, so Iontos doesn't read it all the same way.
- Prose and documents — filings, articles, reports, web pages — are read by a language model that pulls out the entities and relationships described in the text.
- Structured records — registry exports, reference datasets, API responses — are handled by a deterministic path that maps fields directly, without sending well-structured data through a language model.
The second path matters more than it looks: when a source is already structured, mapping it directly is more accurate and far cheaper than asking a model to re-read it — and it removes any chance of a model inventing a detail that the data never contained. Iontos uses the right tool for each kind of source.
Resolution is where the value is
Extraction produces raw entities and facts. Resolution turns them into knowledge:
- The same thing, once. Five sources mentioning "Acme Corp", "ACME Corporation", and a registry number are recognized as one entity — across both public and your private data.
- Disagreements, reconciled. When sources conflict, Iontos reconciles them into a single answer rather than leaving you with contradictions.
- Quality, checked. Built-in data-quality checks catch the gaps and inconsistencies before they reach the graph.
Every fact carries its origin
Nothing in the graph is an orphan claim. Each fact stays linked to the source it was drawn from, so an analyst can always answer the only question that matters in a regulated or high-stakes setting: how do we know this?
Next: the two-graph model — how public consensus and your private knowledge live in one view without ever mixing.