Five million pages of discovery, read on a box you own — on a machine that never touched the internet.
The average big case is now five million pages. Cloud AI can read them — but only by copying the privileged file to a vendor's servers. Probatum reads them locally on hardware you own, inside your building, with the network cable pulled.
100% Free Software. No $6.5M fleet doors. No license fees. No support contracts. Run Probatum locally on a single workstation or cluster.
Every prompt into cloud AI is a privileged file leaving the building.
A modern big-litigation matter averages five million pages of pleadings, discovery, medicals and produced comms. Case files, medical records, settlement numbers — cloud AI copies every one of them to a vendor's servers the moment they're prompted. Your confidentiality duty doesn't outsource, but the vendor's terms of service now sit inside your ethics posture.
Probatum solves this by running 100% locally on hardware inside your building. Zero cloud egress. Zero data leak. Forensic chain-of-custody logging standard.
| Privileged data location | Your local machine |
| Software license | Apache 2.0 Free |
| Breach surface | Zero (air-gapped) |
| Network requirements | None (cable pulled) |
Drop five million pages on Friday. This is waiting at 8am Monday.
On the fictional sample matter Morgan v. Lakeshore Freight & Logistics, Inc., Probatum read the full production — pleadings, discovery, 4.2M produced pages, 240,000 medicals, 96,000 emails — extracted the timeline, entity graph, and issue flags, and staged the drafts. Machine flags, human decisions: nothing files itself.
Among the flags: a produced email chain that contradicts the defendant's sworn interrogatory answer on dock-plate maintenance. Every flag cites the source Bates page it came from.
| Network egress | 0 Bytes |
| Cloud APIs called | None |
| Source-file integrity | SHA-256 verified |
| Third parties holding data | Zero |
| Chain-of-custody log | Per file, per run |
The answer your GC and your malpractice carrier will love.
Every run logs network state and produces a hash-verified manifest of every source file — before and after. No cloud AI, no data processor agreements, no third party to audit. When a client asks "does AI touch my file?", your answer is: yes — on hardware we own, inside our building, that has never been connected to the internet.
Unplug it and the data still never moved.
You don't need 500 GPUs. You need one box.
The minimum hardware to run Probatum locally:
AMD Ryzen AI Max+ 395, 128GB unified memory, runs 128B+ parameter models locally. ~$2,959. Buy it yourself, from them, not from us.
The software is the gift. Download Probatum.
Apache 2.0. No account, no license fee, no support contract. Clone it, change it, use it.
Frequently Asked Questions.
One builder. No sales team. Honest floor, zero oversell.
I'm MadBrad Smith. Thirty-four years running film sets taught me one thing that turned out to be the whole job: span of control. Nobody — no commander, no first AD, and no AI orchestration layer — runs more than five direct reports well. I build AI systems the way I ran sets: narrow specialists, clean handoffs, one outcome. Probatum is that discipline in code.
The software is the gift. The hardware spec is the honest answer to "what do I need." The engine underneath — Developum AI — is mine. The catalog it powers lives at KissMySaaS.Skin.
Here's what it does, here's what you need to run it, here it is.
"Here's what it does, here's what you need to run it, here it is. The software is the gift. The hardware spec is the honest answer."— MadBrad Smith