BRIDGEInnovations
Under the Hood

Not a document storage tool. A knowledge compiler.

Wiki is built on a "LLM-as-compiler" pattern — raw documents go in, structured knowledge comes out. Everything runs locally. No cloud. No vector database. Just folders, prompts, and a local AI that understands your practice.

🖥️

Runs Entirely On-Premise

Wiki runs Ollama locally on your practice's hardware — no cloud API, no external subscription. OCR for scanned images uses Apple Vision on macOS with an automatic fallback. Every byte of your practice data stays on your network.

⚙️

LLM-as-Compiler

Each folder in your vault maps to a tailored prompt. Drop a PDF, image, or Markdown file in — Wiki reads it and compiles a structured wiki page automatically. The output is plain Markdown you can open, edit, or audit in any text editor.

💬

Staff Q&A Engine

Staff type a question in plain English. Wiki runs a keyword search across compiled pages, gathers the top matches as context, and streams a direct answer from the local model. Every answer is saved as its own Q&A page — your knowledge base grows with every question.

🔍

Gap Detection (Lint)

On demand, Wiki reads your entire compiled knowledge base and generates a gap report — identifying which protocols, procedures, or policies are missing or thin. Run it after each batch of new documents to keep your manual complete.

📂

Auto Inbox Classifier

Upload anything to the inbox. Wiki reads the document content — including OCR'd images — and suggests the right destination folder with a confidence level. High confidence routes automatically; low confidence flags for your review.

🌐

Staff-Facing Published Site

One click publishes your compiled knowledge base as a fast static site served on your office network. Staff access it from any device without logging in. Your office name is injected automatically — no HTML editing required.

No vector database. No cloud. No black box.

Most AI knowledge tools rely on cloud APIs and vector embeddings you can't inspect or control. Wiki uses repeatable prompts on local models — every output is predictable, auditable, and fully yours. The entire system is folders and markdown files you can open in any text editor.