Group your folders into named vaults, index everything from PDFs to scanned pages with on-device OCR, then ask questions and get answers grounded in your own documents — every one cited with page numbers — from local Gemma models on your Apple Silicon GPU. Put a vault in Legal/Contracts mode, or hand it to Claude Code over a local MCP server. No upload, no account, no cloud. A vault for research that can't leave the room.
Silo Systems builds infrastructure for people who would rather not negotiate with their own software. SiloVault is the same discipline, pointed at one job — turning folders full of documents into a research assistant that answers with citations — done entirely on the machine you already own. Your case files never become someone else's training data, and your archive never lives on someone else's server.
Indexing, search, chat, OCR, transforms, and the MCP server all run on your Mac through a bundled local runtime on your Apple Silicon GPU. The Gemma chat and embedding model weights download once from ModelScope or Hugging Face after onboarding, then SiloVault works offline for the lifetime of the app — in a SCIF exactly the way it works at your desk.
No analytics SDK, no telemetry, no account, no crash reporter, no external API. The bundled FastAPI backend binds strictly to loopback (127.0.0.1:8765) as on-device IPC — it is not a network service. The documents you index, the questions you ask, and the answers you get stay in your Application Support container, never transmitted anywhere.
Questions are answered locally by Gemma models through Apple MLX, grounded in your own documents — and every answer cites its sources with page numbers, so you can check the work. The research assistant a cloud service would charge you a subscription for — and feed your files into — is computed on hardware you already own and kept there.
Point SiloVault at the folders you choose and group them into named vaults — one per case, project, or subject. PDFs, Office documents, HTML, CSV, Markdown, and plain text parse locally, and scanned PDFs and images are read with on-device Apple Vision OCR. Nothing is uploaded; nothing needs a connection.
Hybrid search — SQLite FTS5 keywords fused with local vector similarity via Reciprocal Rank Fusion — pulls the right passages, and a local Gemma model on Apple MLX writes the answer, grounded in your documents. Every answer cites its sources with page numbers, so you can verify it against the file — entirely on-device.
Turn retrieved context into a Summary, FAQ, Timeline, or executive Brief. Put a vault in Legal/Contracts mode for risk flags, clause redlines, and one-click PDF export — an informational document review, not legal advice. Or expose the vault to Claude Code, Codex, and Cursor over a local MCP server that speaks stdio only and opens no port.
Group folders into separate research collections — one vault per case, project, or subject — and search each independently. A vault is a boundary: queries against one never bleed into another.
Chat with local Gemma models through Apple MLX, grounded in your documents. Every answer cites its sources with page numbers — no hand-waving, no hallucinated references, no cloud round-trip.
PDF, DOCX, PPTX, XLSX, HTML, CSV, Markdown, and plain text parse locally. Scanned PDFs and images (PNG, JPEG, TIFF, HEIC) are read with on-device Apple Vision OCR via the bundled silovault-ocr CLI — no cloud OCR.
SQLite FTS5 keyword search fused with local vector similarity via Reciprocal Rank Fusion, so exact terms and fuzzy meaning both land — and the right passage surfaces either way.
Turn retrieved context into a Summary, FAQ, Timeline, or executive Brief — structured documents generated on-device from your own sources, ready to hand off.
A per-vault preset for reviewing agreements on-device: risk flags, clause extraction, a risk-scoring dashboard, negotiation-playbook comparison (built-in NDA/SaaS/M&A playbooks), clause-level compare/redline, and one-click PDF export. Informational document review only — not legal advice, and no attorney-client relationship is created.
Expose a vault to Claude Code, Codex, and Cursor through a local Model Context Protocol server (ohm-backend --mcp). It speaks stdio only — no open network port, no remote connections.
A SwiftUI host boots a bundled FastAPI backend on loopback and renders the vault in a WKWebView from that same origin. No browser, no server, no port open to your network.
Hosted research assistants and contract-review platforms rent you the workflow by the seat and run your documents through their servers. SiloVault is a single purchase, on your Mac, with unlimited local indexing, search, chat, OCR, transforms, and review — because it all runs on hardware you already own, on an MIT-licensed core. Buy once, keep the archive.
Coming to the Mac App Store ⌘SiloVault is in pre-release for the Mac App Store. Join the waitlist and we'll send a single email when it goes live — no marketing list, no newsletter, no follow-up campaigns.
