A desktop app that reads your documents, files them where they belong, and finds anything — documents and photos — from a plain-English description. Everything runs on your machine. Nothing is uploaded, ever.
Four things, each reachable by just talking to it.
It reads every file, builds a knowledge base from the contents, and proposes where each one belongs — with a reason. Nothing moves until you click Apply, and every move can be undone.
Answers cite the exact file and page. If your library doesn't cover the question, it says so — it never invents an answer.
“A man lifting a baby.” “Sunset on a beach.” No tags, no filenames — it looks at the pictures themselves.
It lives in the system tray and watches your Downloads folder. New documents get read and filed automatically — even with the window closed.
One installer, one guided first run. Internet is needed once, to fetch the AI models — after that it's fully offline.
Grab VectorVault-0.1.1-win64.msi above. It installs per-user — no admin rights needed — and adds Vector Vault to startup so it can watch your folders from login.
Windows will warn because the installer isn't code-signed yet. Click More info → Run anyway. You can verify the download against the checksum below first.
On first launch it detects your GPU and RAM, recommends a language model that actually fits your hardware, installs Ollama, and downloads the model with a progress bar (1–9 GB depending on your machine). Then pick the folders to watch — done.
Why the SmartScreen warning? Code-signing certificates require paid identity verification, which a v0.1.0 doesn't have yet. The installer is built from the public source in this repository — you can audit it, build it yourself, or verify the checksum below.
This app moves your files and reads your private documents, so the safety rails are architectural, not promises.
When it organizes a folder, the model emits a plan — data, not shell commands. It cannot run rm, cannot mangle a filename, cannot escape the library folder. You see the whole plan before a single byte moves, and it executes through an engine that logs every move before making it.
If nothing relevant is found in your library, the language model is never even called. Your documents are the only source of truth — an assistant that invents the contents of your medical records is worse than none.
Colliding filenames get a suffix, never replaced. Every filing operation is written to an undo log before it happens, so even a crash mid-move loses nothing.
The language model (via Ollama), the document embeddings, and the photo search all run locally. After the one-time model download, the app works with the network cable unplugged.
The academic classifier ships with a sample B.Tech CSE curriculum. Edit
curriculum.yaml in the install folder to make it file coursework under
your subjects — it rebuilds its knowledge base automatically on the next start.
current_semester: 5
semesters:
- number: 5
subjects:
- name: Operating Systems
code: CS301
description: >
Processes and threads, CPU scheduling, semaphores, deadlock,
paging, page tables, TLB, virtual memory, thrashing...
topics: [paging, deadlock, semaphore, tlb, page fault]Write descriptions in the vocabulary a real document would use — “paging, TLB, page faults” beats “students will learn about memory management.”
Verify your download (PowerShell):
Get-FileHash .\VectorVault-0.1.1-win64.msi -Algorithm SHA256
# Expected:
# 919BA63DB415A2BF2485CB215E2ED1E0D1ABA63B3BDF686764D90170F1BD4398