Skip to content
Devora Labs
Infrastructure

SkyVault

SkyVault is a self-hosted alternative to consumer cloud storage, built for people and teams who want their files on their own hardware without giving up the intelligence that makes cloud storage useful. It combines a storage layer, an AI understanding layer and a security layer into a single system.

Project detail

Type
Personal Project
Category
Infrastructure
Year
2025
Technology
PythonFastAPIPostgreSQLpgvectorTesseract OCRSentence TransformersDockerReactTypeScriptWireGuard
Problem

Self-hosted storage forces an uncomfortable trade-off. You get ownership and privacy, but you lose everything that makes hosted cloud storage usable: you can only find a file if you remember where you put it, scanned documents are invisible to search, duplicates accumulate silently across years of uploads, and access control is usually a single shared password. Meanwhile a NAS exposed to the internet is a standing target, and most home and small-office deployments have no way to notice they are being probed.

Solution

SkyVault layers intelligence and security on top of ordinary file storage. Uploaded documents and images pass through an OCR and embedding pipeline, so their contents — not just their filenames — become searchable in natural language. Perceptual and semantic comparison surfaces near-duplicate files that byte-level hashing misses. Access is governed by role-based permissions rather than a shared credential, and an anomaly-detection layer monitors access patterns and authentication attempts, flagging behaviour that departs from the established baseline. Networking is closed by default, with encrypted remote access rather than open port forwarding.

Main features

What the system does

01

AI-powered semantic search

Find files by describing them. Queries run against embedded document content, not just filenames and tags.

02

OCR pipeline

Scanned documents and images are read on ingest, making previously opaque files fully searchable.

03

Semantic deduplication

Detects near-duplicate documents and visually similar images that identical-hash checks never catch.

04

AI intrusion detection

Behavioural monitoring of access and authentication patterns, with alerting on anomalous activity.

05

Role-based access control

Granular per-user and per-folder permissions, replacing the shared-password model typical of home NAS setups.

06

Secure networking

Encrypted remote access with no exposed ports, so the system is reachable without being publicly attackable.

Outcome

What it demonstrates

Stated as capability rather than invented metrics — we only publish numbers we can stand behind.

Search

Natural-language retrieval across scanned and text documents

Storage

Near-duplicate detection beyond exact-hash matching

Security

RBAC, anomaly detection and closed-by-default networking

Next step

Have a project like this in mind?

Book a free consultation and we'll walk through the approach, the scope and what it would realistically take.

Prefer to just message? Chat with us on WhatsApp.

Chat on WhatsApp