Where it runs
Your hardware, on-premise. Single-tenant. The core pipeline is air-gappable.
BDS ingests, searches, and builds an entity knowledge graph over large document sets — OCR, semantic search, and named-entity analysis — using AI models that run locally on your own machine. On the core pipeline, no document ever leaves your infrastructure.
The short answer to the three questions a security reviewer asks first.
Your hardware, on-premise. Single-tenant. The core pipeline is air-gappable.
Local disk only — the source documents, the search index, and the entity graph.
Loopback by default; any access beyond the host machine is firewall-gated.
Your documents are processed on your own machine — nothing is sent to an external AI provider.
Text understanding, vision/OCR, embeddings, and entity extraction all run on models hosted on your own GPU. No third-party AI service receives your documents or prompts.
text · Qwen3.8-27B-Instruct vision/OCR · Qwen3-VL-8B embeddings · BGE-large-en-v1.5 entities · GLiNER
Source files, the OCR'd text, the vector search index, and the entity knowledge graph are written only to disk on the host machine. Nothing is uploaded to any cloud store.
The vector-database component's usage telemetry is disabled at startup. The core pipeline makes no outbound calls, so it can run fully air-gapped.
An optional module looks up public open-source intelligence on entities BDS finds. It is the one part of BDS that reaches third-party sources on the internet — disclosed as such — so a deployment that uses it is no longer air-gapped. It is invoked deliberately and separately from the core ingest / search / graph pipeline, which makes no outbound calls; only the entity identifiers you select are sent, never your documents.
localhost (127.0.0.1) only — reachable solely from the machine it runs on. A deployment that installs BDS and changes nothing is already closed to the surrounding network.Exposing it to other machines is a deliberate change you make and place behind your firewall. This is a simplicity choice, not a shortcut: there is no separate identity store to stand up, breach, or leak credentials from. Access reuses the perimeter controls you already operate — BDS is reached the same way as any other internal service.
What BDS delivers, and what your environment provides.
Before an install is scheduled, you complete a short security attestation — host on a segmented / firewalled network · OS patched · physical & admin access controlled. Disk encryption on the data drive is strongly recommended, though not required to schedule. The hardening guide defines what each item means in practice.
Because inference and storage are local, sensitive or privileged documents never transit a third party. BDS can be deployed on an isolated network with no internet access.
Access is governed by your existing firewall and network segmentation. The default configuration is closed — the server is reachable only from itself until you decide otherwise.
One trusted deployment — no user-management overhead and no credential store to maintain.
BDS is a self-hosted document-intelligence system: all AI runs locally, all data stays on your hardware, and access is loopback-closed by default and controlled by your network perimeter.