Turn documents into answers. Keep every byte on your network.

Selvo Lens is an enterprise RAG platform that runs entirely on your infrastructure. Upload PDFs, Excel, Word, and scanned documents — then ask questions in natural language.

Air-gappable·GDPR built-in·Self-healing·50+ enforced invariants

Three steps to document intelligence

No cloud accounts. No API keys. Docker Compose up and go.

yourcompany.selvolens.com

What does the liability clause say in the Acme contract?

According to the Acme Corp Master Services Agreement (Section 8.2), liability is capped at 12 months of fees paid. Neither party shall be liable for indirect, incidental, or consequential damages.

Acme_MSA_2024.pdf — p.14 Acme_Amendment_Q3.pdf — p.3
0.94Qwen-2.5-32Bcontent
01

Upload your documents

PDF, Excel, Word, CSV, images — including scanned documents with automatic OCR. Multi-sheet Excel gets agentic sheet selection.

02

Ask in natural language

Content questions get hybrid search + LLM synthesis. Analytical queries route through sandboxed Pandas for deterministic answers.

03

Get cited, governed answers

Every response includes source citations, confidence scores, model transparency, and an immutable audit trail.

Built for environments where cloud is not an option

Security, governance, and compliance are foundational — not add-ons.

Air-gapped deployment

The LLM, vector database, and all user data stay on your servers.

No internet connection required after initial Docker deployment. Suitable for classified environments, defense networks, and any infrastructure with strict network isolation.

Zero cloud dependencySingle GPU hostDocker Compose
terminal
# Your data never crosses your firewall
$ docker compose up -d
✓ frontend running :3000
✓ backend running :8001
✓ chromadb running :8000
✓ vllm running :8080
# Network: internal only. No egress.
Deterministic analytics

Numbers you can trust.

Analytical queries route through sandboxed Pandas code execution. The LLM generates code against your data schema — no hallucinated numbers.

PandasDeterministicAuditable
Full governance

50+ invariants. Immutable audit trail.

Policy signing with hash-chain integrity. Department-scoped RBAC with 4 roles. Consent management, data retention, right to erasure, and breach detection.

RBACGDPRAudit ledgerConfidence gatingFail-closed

Self-healing infrastructure

Auto-rebuilds vector index from immutable SQLite ledger

Confidence gating

Below-threshold responses abstain — fail-closed by design

Model agnostic

Swap LLM, embedding model, or cross-encoder via .env

Response transparency

Every answer shows model tier, confidence, and source citations

Cross-language queries

LLM-based reranking fallback for any query language

On-premise OCR

Tesseract 5 for scanned PDFs — even blurry documents

Built for industries where cloud AI is banned

Your sector can't use ChatGPT or Claude due to compliance risks. Selvo Lens gives your teams the same power — without the exposure.

Legal & Law Firms

Query discovery documents and case files without violating attorney-client privilege.

Financial Services

Analyze internal audits, KYC documents, and market reports under strict data sovereignty rules.

Defense & GovTech

Air-gapped deployment for classified and sensitive mission data. No internet required.

Manufacturing & R&D

Protect intellectual property and blueprints from being used to train public models.

Cloud RAG vs. Selvo Lens

See why regulated industries are choosing on-premise over cloud AI.

Data Privacy

Cloud RAGShared with provider
Selvo LensZero-leak / On-premise

Math & Analytics

Cloud RAGHallucinates on numbers
Selvo LensDeterministic Pandas routing

Compliance

Cloud RAGHard to audit
Selvo LensImmutable ledger & GDPR built-in

Deployment

Cloud RAGSubscription / OpEx
Selvo LensYour infrastructure / Predictable cost

Network Requirement

Cloud RAGAlways online
Selvo LensAir-gappable

Confidence Handling

Cloud RAGAlways answers (may hallucinate)
Selvo LensFail-closed with confidence gating

Audit Trail

Cloud RAGLimited or none
Selvo Lens50+ invariants + hash-chain integrity

Intelligent query routing

Every query is classified and routed to the right engine. Content questions get semantic search. Analytical questions get deterministic code execution.

Content

"What does the contract say about liability?"

Hybrid vector + BM25 search with cross-encoder reranking, then LLM synthesis with cited sources.

Analytical

"Average revenue by region for Q4"

LLM generates Pandas code against your data schema. Sandboxed execution returns deterministic results.

Filter & Lookup

"Show all rows where status is Active"

Direct DataFrame filtering and targeted record search. No LLM hallucination on structured data.

Executive Summary

"Give me an executive summary"

Multi-sheet LLM synthesis across entire documents. Produces structured overviews with key findings.

Metadata

"How many documents are uploaded?"

Collection-level metadata queries answered directly from the document ledger.

Cross-language

Queries in any language

LLM-based cross-language reranking fallback when embeddings cannot handle the query language.

Hybrid search that adapts to each query

Every retrieval runs through dual engines with adaptive weight fusion - because metadata lookups need different retrieval weights than open-ended content questions.

Vector search - Semantic similarity via sentence-transformer embeddings in ChromaDB
BM25 keyword search - Lexical matching for exact names, IDs, and codes
Adaptive fusion - Weights shift per query type - 80% vector for content, 95% BM25 for metadata
Cross-encoder reranking - Fine-grained relevance scoring on fused results
Confidence floor - Results below threshold are rejected - never guesses
Adaptive weight fusion by query type
ContentBM25 20% · Vector 80%
AnalyticalBM25 60% · Vector 40%
FilterBM25 70% · Vector 30%
LookupBM25 80% · Vector 20%
MetadataBM25 95% · Vector 5%
BM25 (keyword) Vector (semantic)

Four containers. One GPU host. Complete sovereignty.

The entire stack runs on a single machine via Docker Compose. No cluster, no Kubernetes, no cloud.

Docker Compose · Internal Network
Frontend
Next.js · :3000

Browser-based UI for document upload, querying, and admin dashboard.

Backend
FastAPI · :8001

Query routing, classification, hybrid search, analytics engine, GDPR, governance.

ChromaDB
Vectors · :8000

Vector embeddings store. Reconstructible from upload ledger if corrupted.

vLLM Inference
GPU · OpenAI-compatible

Local LLM inference. Model-agnostic - swap via .env config per deployment.

Air-gapped Self-healing GPU-accelerated 4 uvicorn workers

Multi-format document ingestion

Upload anything. Scanned documents are OCR-processed automatically. Excel files get agentic sheet selection.

PDF
Text + scanned OCR
Excel
Multi-sheet, agentic
Word
Paragraphs + tables
CSV
Auto-encoding
Images
OCR via Tesseract 5
Text/MD
UTF-8

What teams are saying

From teams in regulated industries running Selvo Lens in production.

Didn't expect it to work on our old scanned PDFs but it actually pulled the data correctly. The OCR handled documents our previous tools couldn't touch.

M.K.

Legal

IT approved it fast because nothing leaves our network. That never happens. The air-gap capability was the deciding factor for our compliance team.

R.S.

Compliance

We finally have an AI tool that can answer 'What was the total spend last quarter?' without making up numbers. The Pandas routing is a game changer.

J.P.

Financial Operations

Runs on commodity hardware

The entire stack - frontend, backend, LLM inference, and vector database - runs on a single GPU host via Docker Compose. No cluster required.

View pricing details
Minimum Requirements
CPU
8 cores / 16 threads
RAM
32 GB
GPU
24 GB VRAM (NVIDIA)
Storage
256 GB NVMe SSD
Recommended
CPU
8 cores / 16 threads
RAM
32 GB DDR4/DDR5
GPU
RTX 5090 / RTX 6000
Storage
512 GB NVMe SSD

Common questions

Answers for CTOs, CISOs, and compliance officers.

Stop the shadow AI problem

Give your teams the AI-powered document intelligence they need — on infrastructure you control, with governance your compliance team will approve.