Document search
Ask questions across millions of pages — PDF, Word, spreadsheets and scanned images via OCR. Every answer cites file and page.
CAPABILITIES
Everything runs behind your firewall.
Hybrid RAG engine
Vector similarity for meaning, SQL for exact fields and dates, a knowledge graph for relationships — merged into one grounded answer.
Custom chat models
Local models tuned on your terminology, formats and policies. From 7B on a single workstation to 70B on a GPU node.
Specialized agents
Multi-step agents for complex tasks: contract review, compliance checks, report assembly — each step logged and auditable.
ARCHITECTURE
One pipeline, three indexes.
Documents are parsed, OCR’d and chunked once — then indexed three ways so every question finds the right kind of evidence.
RUNS ENTIRELY INSIDE YOUR NETWORK
Your documents
PDF · DOCX · XLSX
TIFF · PNG scans
Ingestion
parse · OCR
chunk · embed
Vector tablesVectorANN
SQLSQLEXACT
Knowledge graphGraphLINKS
Hybrid retriever
rank · merge
deduplicate
Local LLM
answers + citations
agents · tools
Vector — finds passages by meaning, even when wording differs across documents.
SQL — exact filters over extracted fields: dates, amounts, parties, statuses.
Graph — entities and relations, so amendments, parties and clauses stay connected.
SECURITY
Air-gapped by design, not by policy.
There is no cloud fallback and no telemetry endpoint to disable — the software has no route out of your network to begin with.
- Zero external calls. Inference, indexing and OCR run on your GPUs — or on a pre-configured appliance we ship.
- Role-based access + audit log. Every query, source and answer is recorded and attributable.
- Offline updates. Models and knowledge bases arrive as signed bundles you install on your schedule.
- Your data stays yours. Nothing is used for training elsewhere — there is no “elsewhere”.
YOUR NETWORK
Documents
file server · DMS
Indexes
vector · SQL · graph
Local LLM
GPU node / appliance
Your team
browser · LAN only
AIR GAP
Cloud APIs
no route out
DEPLOYMENT
From audit to answers in four weeks.
-
01 · DAYS 0–3
Scope
We audit your document estate, tasks and hardware. You get a sizing plan — GPU node or shipped appliance.
-
02 · WEEK 1
Deploy
Install on your network — no inbound or outbound internet required. Integration with AD/LDAP and file shares.
-
03 · WEEKS 2–3
Index
Ingestion pipeline parses, OCRs and indexes your documents into vector tables, SQL and the knowledge graph.
-
04 · WEEK 4
Specialize
We tune chat models on your terminology and configure agents for your workflows. Your team goes live.
KNOWLEDGE BASES
Domain expertise, pre-loaded.
Curated, versioned and citable corpora that ship with your deployment — updated through the same signed offline bundles.
Law
Statutes, case law, contract doctrine and regulatory texts.
2.1M SOURCES
Medicine
Clinical guidelines, drug interactions, coding and protocols.
870K SOURCES
Finance & tax
Accounting standards, tax codes and reporting frameworks.
640K SOURCES
Engineering
Industry norms, standards and technical documentation.
1.3M SOURCES
Your field
We build and maintain a corpus for your domain, to order.
BUILT TO ORDER →
CONTACT
Talk to an engineer.
No sales script. Describe your documents and tasks — we’ll tell you what hardware it needs and what a pilot looks like.
- → reply within one business day
- → NDA on request
- → on-site or air-gapped pilot available