Home

Sift Documentation

Welcome to the Sift documentation. Sift is a privacy-first, local-only clinical intelligence engine designed for modern medical offices. It monitors local folders for incoming clinical data — EPIC exports, FHIR JSONs, HL7 messages, and scanned PDFs — and synthesizes them into actionable clinical summaries using local LLM inference. All processing happens on the local machine. No data ever leaves the office network.

Vision

To empower medical office staff and clinicians with zero-click, privacy-first clinical intelligence that works automatically in the background — turning fragmented data formats into clear, actionable summaries without touching a cloud service.

Mission

Sift removes the friction between raw clinical data and usable insight. By combining a Rust-native watchdog, a Node.js orchestration layer, and local LLM inference, Sift gives healthcare teams immediate access to synthesized patient records — structured or not — directly on their existing Windows workstations.

Quick Links

• Quick Start
• Installation
• Architecture Overview
• Configuration Guide
• Build and Release
• Security and Compliance
• Contributing

Project Overview

Sift is a local-first desktop application built on a polyglot sidecar architecture:

• Automatically monitors a designated folder for new clinical files
• Detects and parses FHIR JSON, HL7 v2, and PDF formats
• Calls a local OpenAI-compatible LLM (Ollama, LM Studio, llama-server) for clinical narrative generation
• Stores summaries, confidence scores, and source previews in a local SQLite database
• Presents findings in a clean React dashboard with printable report output

Key Features

• Zero-Cloud Architecture: All processing happens on-device. No data is transmitted to external services in default configuration.
• Automated Ingestion: Folder watcher detects new files and begins processing immediately with no user action required.
• Multi-Format Support: Handles structured FHIR JSON, pipe-delimited HL7 v2, and unstructured clinical PDFs.
• Local LLM Inference: Works with any OpenAI-compatible local server (Ollama, LM Studio, llama.cpp).
• Confidence Scoring: Every synthesized summary is stored alongside a confidence score and heuristic fallback.
• Single-Binary Distribution: Ships as a self-contained Windows NSIS installer bundling Tauri, the Node.js sidecar, and optionally llama-server.
• Printable Reports: The React UI includes Tailwind print classes for professional clinical document output.
• Audit Trail: All document records are persisted in SQLite with timestamps and error states for review.

Architecture Summary

Sift is composed of three cooperating processes:

Layer	Technology	Role
Host	Rust / Tauri	Window management, system tray, folder watcher, sidecar lifecycle
Orchestrator	Node.js / Express	REST API, ingest pipelines, SQLite, LLM client
UI	React / Vite / Tailwind	Dashboard, document viewer, settings, print output

See Architecture for the full component breakdown and data flow.

Repository Structure

sift/
├── .github/
│   ├── workflows/ci.yml       -- CI: backend tests, frontend tests, cargo check
│   └── workflows/release.yml  -- Release: Windows NSIS installer on version tag
├── docs/                      -- Source documentation
├── frontend/                  -- Vite + React + Tailwind UI
├── backend/                   -- Node orchestrator (Express, SQLite, ingest, LLM client)
├── src-tauri/                 -- Tauri host: tray, folder watch, sidecar lifecycle
├── samples/                   -- Sample FHIR, HL7, and PDF files for manual testing
├── img/                       -- Brand assets
├── sift.mjs                   -- Developer CLI
└── data/                      -- Local SQLite (sift.db), created at runtime

---

Report Bug | Request Feature | Contributing Guidelines

© 2026 Sift / fleXRPL. All Rights Reserved.