Retrievers
A retriever reads the reasoning graph on behalf of a query. Rather than one general search, Virgo ships specialized retrievers, each a graph traversal built for a particular kind of question, and routes every query to the retriever built for it.
The built-in retrievers
| Retriever | Answers | Reads |
|---|---|---|
| Context | How a system works | Architecture, logic, and how the pieces connect |
| Causal | Why it changed | Decisions, incidents, and fixes over time |
| Learning | Team conventions | Patterns and prior learnings, recalled before code is written |
Permission-aware by default
Retrievers enforce your access control during retrieval. Results are filtered at query time, so people and their agents only ever see data they already have access to. There is no separate sync to keep in step.
Writing your own
When your domain needs a traversal the built-ins do not cover, you can write one. A new retriever is a new traversal over the graph, and you can call it from your endpoint as soon as it is registered. See Custom retrievers.