How it works
From approved sources to permission-filtered answers.
BlackBox Node is designed around read-only ingestion, local indexing, local model runtime selection, matter and case permissions, and a local audit trail.
Source access
Connect approved data read-only
The ingestion direction is read-only. Source systems should remain authoritative, and connector scope should be approved before any source is indexed.
Retrieval boundary
Filter context before the model sees it
Search results should be limited by matter, client, case, role, and user permissions before any local model receives context.
Answer review
Grounded output with citations
The product direction favours answer support, references, citations, and professional review rather than unsourced automation.
Query path
Permission-filtered retrieval happens before local generation.
This is the high-level query flow the public website can explain without becoming an appliance control plane.
-
01
User selects scope
The user works inside an authorised matter, client, case, or permitted knowledge scope.
-
02
Retrieve local chunks
The appliance searches local indexes built from approved read-only sources.
-
03
Apply permissions
Matter and case permissions remove content the user should not see.
-
04
Generate locally
A local model runtime produces a grounded draft from permitted context.
-
05
Show citations
The UI presents answer support so professionals can review source material.
-
06
Write audit event
The appliance records non-sensitive query metadata in the local audit trail.
Public website boundary
This site explains the product. It is not the appliance.
This public website is product information only. Do not submit confidential client, patient, case, investigation, regulated, or commercially sensitive data through the public website.