Launching today
IPVault lets you use AI on confidential scripts, patents, business plans, and trade secrets without teaching it your ideas. Your document is split into masked pieces across several AI models, so no single AI ever sees your whole work, and none of it is ever used to train AI. IP-safe AI for writers, inventors, founders, and law firms.
This looks really useful for anyone worried about AI leaking sensitive work. One thing I'd love to see is a clear audit log or activity history so users can review which model handled which chunk and when. That kind of transparency would make it way easier to trust the split-routing setup and explain the security model to clients at a law firm.
@yasemineryitn Great feedback, and the good news is: we already do this it's the Trust Layer we shipped. That reviewer is describing your differentiator almost word for word. Here's exactly where it lives today:
Trust → Sources: your document, chunk by chunk in order. Open any chunk and it shows precisely which AI handled it and what it sent back, with the sensitive ones marked "kept on your device." That is "which model handled which chunk."
Trust → Proof: the tamper-evident, timestamped chain of custody, including the full step-by-step record (every dispatch and response, hash-chained, with times). That is the "when" and the "audit log."
Pieces: the per-piece table (piece, task, AI model, status, prompt hash).
So a law firm can already review which model saw which chunk, and prove nothing was altered. It's genuinely a strong "explain it to clients" story.
Would love to see a way to mark specific sections of a document as extra sensitive so they get split into even smaller pieces or routed to a different model pool. That would give writers and patent attorneys more control when one paragraph carries the actual invention claim.
@hakanbaayaqrjy great call Hakan, we had this turned off for launch, we turned it back on. You are a smart man!