Evaluating infrastructure usually means taking an architecture diagram on faith. Someone draws five boxes, joins them with arrows, and you are invited to believe that the arrows mean something.
We have published an interactive map of SWIRL Enterprise 5.0 that works the other way round. Thirty-seven components, forty-seven connections, and every one of them traced to a real part of a running installation before it was drawn. You can drag it, zoom it, and click any box to see what it does and what it replaces.
It is also running, right now, behind the headline on our home page.
The arrows are the point
Most diagrams label an arrow with its direction. Ours are labelled with the transformation that happens along them. Not "sends to" but "query template to live SQL". Not "returns" but "source responses to common result shape". Once you label arrows that way, the diagram stops being decoration and starts being an argument about how the system behaves.
Colour carries meaning too. A box is a source of truth, a derived index, transport, processing, a service, one of your own systems, or an optional capability. If a box is a derived index, it can be rebuilt rather than restored, and your backup plan just got simpler. That is a real operational fact, encoded in a colour.
Four journeys you can replay
The map has a Traces panel. Pick a journey and it steps through the system one connection at a time, dimming everything else and explaining what happens at each step:
- A federated search. One question reaches every connected system at the same moment.
- Recall of a document already read. How meaning-based recall surfaces a document when the wording is completely different.
- An AI answer. How sources are chosen, budgeted and turned into a cited answer.
- An agent searches. The same pipeline, reached over the Model Context Protocol.
What the shape gives away
Four things are easier to see in the map than to say in a sentence.
Your data does not move. Follow any arrow from the connectors out to Microsoft 365, a database or an existing index. It goes out, and results come back. There is no arrow carrying your content into a SWIRL index, because there is no such index. No crawl schedule, no re-index window, no second copy of sensitive content to secure and audit separately.
Permissions are enforced where they live. Every connector arrow carries the signed-in user's own credential. Microsoft and your databases apply their own rules before SWIRL sees a single row, so SWIRL never has to reproduce your permission model, and cannot get it wrong.
One scale is what makes a merged list a ranking. Every system scores its own results on its own scale, and those numbers mean nothing to each other. Watch the re-scoring stage in the trace. Without it, a federated result list is several lists stapled together.
Answers are built from documents, not snippets. In the AI answer trace, look for the step into document fetch. A search result carries a few hundred characters of preview text. SWIRL retrieves the real document and writes the answer from the whole thing. It is the single largest factor in whether an answer is useful or vague.
Have a look
Open the map and follow a search all the way through. If it raises a question about how SWIRL would sit in your environment, that is exactly the right question, and we are happy to answer it against your own stack.