GUIDE (Public)

GUIDE is a monorepo with:

• apps/app (Expo web frontend)
• apps/server (Express backend)

Abstract

Digital mental health (DMH) tools have extensively explored personalization of interventions to users' needs and contexts. However, this personalization often targets what support is provided, not how it is experienced. Even well-matched content can fail when the interaction format misaligns with how someone can engage. We introduce generative experience as a paradigm for DMH support, where the intervention experience is composed at runtime. We instantiate this in GUIDE, a system that generates personalized intervention content and multimodal interaction structure through rubric-guided generation of modular components. In a preregistered study with N=237 participants, GUIDE significantly reduced stress (p=.02) and improved the user experience (p=.04) compared to an LLM-based cognitive restructuring control. GUIDE also supported diverse forms of reflection and action through varied interaction flows, while revealing tensions around personalization across the interaction sequence. This work lays the foundation for interventions that dynamically shape how support is experienced and enacted in digital settings.

Paper: https://arxiv.org/abs/2604.07558
Project page: https://ananya-bhattacharjee.github.io/guide/
Code: https://github.com/StanfordAI4HI/guide-public

Important Disclaimer

This public demo showcases the system’s ability to generate interventions for general stress situations. It is not intended to address safety-critical use cases or complex edge cases.

Interactions are not monitored, and no real-time support is provided. Do not use this tool for crisis, emergency, or high-risk situations.

While some basic safeguards are included, this system does not replace professional care. If you need additional support, please contact a qualified professional or local services.

Research App & Maintenance

This is a research application built to demonstrate the ideas in the GUIDE paper. We will try to keep it maintained, but ongoing maintenance is difficult: the underlying models change constantly, and older ones are frequently updated or deprecated. Expect some drift over time.

Note on image generation: the current code defaults to OpenAI's gpt-image-1 model, whereas the deployed study version used dall-e-3. Output format and behavior may therefore differ between this source and the version used in the study.

Prerequisites

• Node.js 20+
• npm 10+

Quick Start (Install + Run)

git clone https://github.com/StanfordAI4HI/guide-public.git
cd guide-public
npm install
npm --prefix apps/app install
npm --prefix apps/server install
cp apps/server/.env.example apps/server/.env
cp apps/app/.env.example apps/app/.env.local

Set your OpenAI key in apps/server/.env:

OPENAI_API_KEY=your_key_here

Start backend:

npm --prefix apps/server run dev

Start frontend (new terminal):

npm --prefix apps/app run web

Open the URL shown by Expo (usually http://localhost:8081).

Environment Variables

Server (apps/server/.env):

• OPENAI_API_KEY (required)
• PORT (optional, default 8787)
• OPENAI_IMAGE_MODEL (optional, default dall-e-3)
• CORS_ORIGINS (recommended in production; comma-separated allowlist)
• ALLOW_SESSION_READS (optional; default off in production)
• ALLOW_LOG_STREAM (optional; default off in production)
• RATE_LIMIT_ENABLED (optional; default on in production)
• RATE_LIMIT_MAX (optional; default 120)
• RATE_LIMIT_WINDOW_MS (optional; default 60000)

App (apps/app/.env.local):

• EXPO_PUBLIC_API_BASE (default template: http://localhost:8787)

Security Notes

• Keep real secrets only in local env files or your cloud secret manager.
• Do not put real secrets in .env.example or source files.

Contact

For questions, contact: ananyabh@stanford.edu