OpenEdge: Building an Offline First-Aid & Emergency Assistant with Open-Source AI

Description and use

OpenEdge delivers offline emergency triage using open-weight AI models and WebAssembly, eliminating network dependency.

Problem

During natural disasters, connectivity loss, or travel in remote regions, standard cloud-based AI solutions become inaccessible the moment an internet connection drops, making triage advice dependent on network bandwidth, paid API keys, or remote cloud servers.

Outcome

OpenEdge delivers reliable, offline-first emergency triage guidance using lightweight open-weight models, eliminating dependence on internet connectivity and ensuring data sovereignty.

Workflow

• 1. User opens OpenEdge in a modern web browser.
• 2. The application loads a 4-bit quantized open-weight language model (e.g., SmolLM or Gemma 2B) via WebAssembly and WebGPU.
• 3. The model executes constrained decoding (temperature ≤0.2) with prompt-anchored decision trees to generate actionable triage steps.
• 4. Results are displayed to the user without any network requests.
• 5. Service workers and Cache API enable instant offline reloads.

Evidence

During natural disasters, connectivity loss, or travel in remote regions, standard cloud-based AI solutions become inaccessible the moment an internet connection drops.

Source

It operates directly inside any modern web browser to deliver deterministic, step-by-step first-aid guidance completely offline.

Source

Using open weights allows the application to embed model runtimes directly on-device using WebAssembly (Transformers.js / ONNX Runtime Web).

Source