Still Here

Interface
Knowledge Graph

Inspiration

Grief support often arrives all at once, then fades while difficult moments keep coming. We wanted to create something that stays quietly present and offers support when it matters most.

What it does

Still Here is a memory companion that knows when to call. It connects approved memories with upcoming calendar moments - such as interviews, anniversaries, or family traditions - and shares timely encouragement through a gentle voice conversation.

How we built it

We built the experience with Next.js, React, and TypeScript. A LangGraph service evaluates upcoming events, selects relevant memories, applies safety checks, and prepares each outreach. ElevenLabs and Twilio power real-time phone conversations, while SQLite stores profiles, memories, events, and call history.

Challenges we ran into

The hardest challenge was making proactive outreach feel supportive rather than intrusive. We needed strong safeguards around consent, quiet hours, contact frequency, sensitive topics, impersonation, and dependency language. We also had to give the voice agent enough context to feel personal without exposing unnecessary private information.

Accomplishments that we're proud of

We created an end-to-end proactive agent that can independently decide whether, when, and why to reach out. Every call is grounded in user-approved memories, safety-checked twice, and accompanied by an inspectable trace explaining the agent’s decisions.

What we learned

Empathetic AI depends less on sounding human and more on respecting boundaries. Clear identity, consent, restraint, transparency, and the ability to decline are essential - especially when designing for emotionally sensitive moments.

What's next for Still Here

Next, we want to add calendar integrations, richer memory capture, improved personalisation, trusted-family controls, and stronger long-term safety evaluation. Our goal is to make Still Here a dependable companion that supports people without trying to replace those they have lost.