Designed from the ground up for networks that cannot reach the public internet. The browser is the interface. The perimeter is the product.
Deployment
The platform adapts to your network topology — not the other way around.
Deployed entirely within your facility. No cloud connections required. Hardware and OS agnostic.
Operates in fully isolated environments with no physical network path to the internet. Offline-first by design.
Synchronized between distrubted peer-to-peer networks and central command-and-control.
Lightweight deployments for resource-constrained hardware in the field. Designed for intermittent connectivity.
Mapping & Geospatial
Built on Open Geospatial Consortium standards. Serves map tiles, feature data, and spatial queries entirely from within your network.
Implements WMS, WFS, 3D tile, and related standards. Compatible with standard GIS clients without modification.
No client software to install. The full mapping interface runs in any modern browser on your local network.
Tile caches, feature stores, and services operate without upstream connectivity once deployed. Add data in the field.
Data Sharing
Content moves directly between endpoints on the network.
Files.Plus is a browser-native application that synchronises entire folder trees — preserving the full nested structure — between two peers via either direct connections or a proxy. No server ever touches the data. Peers exchange a room link, select local sync folders, and click "Sync Now." The platform handles the rest: scanning, diffing, chunked transfer, hash verification, and conflict resolution — all inside the browser with zero installs.
Synchronises entire directory trees — nested folders, hundreds of files — in a single click. Both peers see the same structure when it's done.
Interrupted mid-transfer? Reconnect and pick up from the last checkpoint. Crash-safe progress persists to IndexedDB every 128 MB.
BLAKE3 hashes identify identical files regardless of timestamps. Already synced? Skipped instantly — no wasted bandwidth.
Every transferred file is verified against its BLAKE3 hash on arrival. Mismatches trigger automatic retries — up to three per file.
No cloud accounts, no servers to maintain, no storage fees. The browser is the entire client — open a URL and start syncing. The pull-based protocol ensures the requesting peer decides what to download; nothing is ever pushed unsolicited.
Chunks write directly to disk as they arrive — RAM usage stays flat at 128 KB regardless of file size, enabling multi-gigabyte video syncs without freezing the browser. NAT traversal via STUN/TURN handles firewalls, VPNs, and symmetric NATs across all Chromium-based browsers.
Encrypted data transfer between endpoints. Direct connections when peers can reach each other, relay-assisted when they cannot.
Native browser API for reading and writing files on disk. Enables streaming writes — no in-memory accumulation.
Rust-compiled WASM module delivers 3–5× faster hashing than SHA-256. Streaming architecture feeds chunks into the hasher alongside disk writes.
Sliding-window congestion control with RTT-based adaptation. Fills the network pipe continuously instead of waiting for one ack per chunk.
The sync engine implements techniques from published delta synchronisation research — production-grade, not theoretical.
Implemented · Dsync (IEEE MSST 2020) · NetSync (IEEE TPDS 2022)
Content-aware boundary detection replaces fixed-size splitting. This solves the boundary-shift problem — insertions and deletions no longer invalidate every subsequent chunk, reducing re-transfers dramatically.
Implemented · Dsync · NetSync
The FastFP algorithm piggybacks on rolling hashes already generated during content-defined chunking, producing a weak fingerprint at 1.5× the speed of Adler32 with equivalent collision resistance. Strong hashes are computed only for weak-hash-matched chunks.
Implemented · Dsync · NetSync
Consecutive weak-hash-matched chunks are merged into a single large block, reducing metadata traffic by up to 80%. Strong-hash verification runs once per merged block — not once per constituent chunk.
Implemented · NetSync
Chunking granularity and compression strategy adjust based on measured network conditions — finer chunks under constrained bandwidth, coarser chunks on high-speed links — to minimise total sync latency.
Production teams exchanging large video rushes, RAW photo sets, and project files between edit suites without uploading to a third-party cloud.
Lecturers and students sharing course materials, datasets, and research outputs across campus networks without institutional file-size limits.
Laboratories synchronising instrument data, simulation outputs, and genomic datasets between workstations — peer-to-peer, no cloud dependency.
Archival teams and digital preservation units moving collections between on-campus storage systems without external transfer services.
AI Solutions
The right models, the right runtime, and the right workflows — all running within the network perimeter.
Pre-validated model packages matched to the deployment hardware. Includes Qwen 3.6, Laguna S 2.1, Ornith, and Kimi K3 — each with quantisation profiles tuned for memory and latency targets on constrained infrastructure.
Automatic runtime matching for the target environment. CUDA-accelerated inference on GPU workstations, Metal-optimised execution on macOS, and portable CPU runtimes for generic hardware — all from the same deployment package.
Production-ready inference pipelines built for constrained environments. Thinking Cap for structured reasoning, Gigatoken for extended-context generation, Ponytail for multi-step orchestration, and KV cache compression for reduced memory footprint on long sessions.