Nothing leaves your network
The models, the weights and the whole pipeline run inside your perimeter. Vesta makes no outbound request at runtime — not for inference, not for licensing, not for telemetry.
Self-hosted · Air-gapped · On your hardware
Your documents become knowledge.
Your data never leaves your system.
Vesta turns your organisation's documents into a connected knowledge base your team can question in plain language — running entirely on your own GPU, on local models, with no internet connection at any point.
No cloud account. No telemetry. No outbound connections — by design.
Built for work that cannot go to the cloud
Some organisations cannot send documents to a third-party model, and no contractual assurance changes that. Vesta is the same class of system, delivered so it never has to.
You own the hardware
Vesta is installed on a server or workstation you already control. There is no per-token bill and no third-party capacity to queue behind.
Answers you can check
Every answer is grounded in your own documents and cites what it drew on, so a reviewer can follow it back to the source rather than take it on trust.
How it works
Four stages, all of them on your machine.
-
1
Ingest
Point Vesta at your documents — PDFs, Office files, scans. Text extraction and OCR run locally, so even scanned material never leaves the box.
-
2
Understand
Local language models read each document, extract the entities and relationships inside it, and build a knowledge graph across your whole library.
-
3
Ask
Your team asks questions in plain language. Vesta searches the graph and the text together, so it can answer things that no single document states outright.
-
4
Verify
Answers arrive with their sources attached. Nothing is a black box you have to accept on faith.
Editions
One product, sized to the hardware you have. Every edition uses the same document representation, so moving up a tier never means re-processing your library.
NVIDIA GPU server
From a single 16 GB desktop card up to a workstation-class GPU. The larger the card, the larger the model Vesta runs — the product is identical.
- Docker Engine + Compose v2
- NVIDIA driver
- Roughly 45 GB of free disk
Apple Silicon Mac
The same Vesta on a Mac Studio or MacBook Pro, running the models natively on Metal through Docker's model runner.
- Docker Desktop with Model Runner
- Xcode Command Line Tools
- 24–32 GB of memory or more
Not sure which fits? Tell us what hardware you have and we will tell you honestly what it will and will not do.
Licensing that respects the air gap
Verified offline
Your licence is a signed file. Vesta checks the signature on your own machine, with a key built into the product. There is no activation server to reach and nothing to phone home to.
Locked to your machine
A licence is issued against your GPU's unique identifier — you send us a fingerprint once, and we send back a licence that runs on that machine.
Updates included
Every release is free while your term is valid, and updates preserve your data. If a term lapses there is a grace period, and your data is never touched.
The full detail is public: Administrator Guide.
Getting started
Three steps, and the first one takes a minute.
1. Fingerprint your machine
Run one small script and send us what it prints. It reads nothing but your GPU's identifier.
2. We issue your licence
We send back a licence locked to that machine, along with download links for the release.
3. Install
Download, unpack, run the installer. The commands are written out in full, with nothing to work out.