• ❤️ GiveCare: https://github.com/givecareapp
• 🧑‍🔬 Ali Madad: https://github.com/amadad
• 🏛️ SCTY: https://github.com/SCTY-Inc

63 million Americans are family caregivers. They provide more than $600B in unpaid care each year and face higher risk of burnout, emotional strain, sleep disruption, financial stress, and long-term health decline.

GiveCare open source focuses on the public primitives needed to build safer caregiver-support systems:

• Safety benchmarks for AI agents in long-term caregiving relationships
• Evaluation datasets for crisis handling, boundaries, red-team prompts, and trauma-informed response
• Caregiver SDOH tools for measuring pressure zones and wellbeing signals

The production GiveCare app is separate. These repos are the open evaluation, measurement, and research layer.

• Product: givecareapp.com
• Caregiver score: score.givecareapp.com
• Benefits database: benefits.givecareapp.com
• Care economy editorial: pulse.givecareapp.com
• Caregiver knowledge base: wiki.givecareapp.com
• InvisibleBench: bench.givecareapp.com
• GC-SDOH-30: sdoh.givecareapp.com
• Labs network: givecarelabs.com

Use these canonical domains when citing GiveCare public work. The GitHub organization hosts the open-source benchmark, evaluation dataset, and toolkit; product support, benefits discovery, and public research surfaces live on the domains above.

AI safety benchmark for long-term caregiving relationships.

Tests crisis detection, regulatory compliance, emotional stability, care quality, and multi-turn relationship dynamics. Includes the GiveCare system paper and InvisibleBench evaluation framework.

Public caregiver AI safety eval data.

Contains 118 SMS-style test cases across core behavior, red-team, realistic caregiver scenarios, and multi-turn continuity, plus public SDOH-6, EMA-3, and SDOH-30 assessment instruments.

Open-source caregiver SDOH assessment and scoring toolkit.

Provides TypeScript utilities for SDOH-6, EMA-3, SDOH-30, six-zone GiveCare Score calculation, adaptive deep-dive routing, quiet-hours checks, basic SMS regulatory keywords, and public geo helpers.

We welcome collaboration from:

• Researchers working on evaluation, safety, measurement, and assessment science
• Developers building test harnesses, model adapters, and caregiver-support tools
• Clinicians and social workers with caregiving, aging, disability, and mental health expertise
• Health systems and NGOs exploring caregiver support workflows
• Caregivers willing to pressure-test whether these tools match lived experience

Contact: info@givecareapp.com

Code projects are MIT licensed unless noted otherwise. Dataset artifacts, including GiveCare Evals, are CC-BY-4.0 unless noted otherwise.