Still Here
Still Here helps people recovering from psychiatric care reconnect with the people they love, even when reaching out feels impossible.
Inspiration
An estimated 1 in 350 people die by suicide within the first three months after psychiatric hospitalization, according to a 2017 systematic review.
For someone leaving psychiatric care, going home can be one of the most vulnerable moments of recovery.
Depression can drain the energy it takes to answer a message, make a plan, or even tell a friend you'd like to see them.
Meanwhile, the people who care about you are living their own busy lives. They may not know you're pulling away or how to help without overwhelming you.
Weeks can pass without anyone spending time together. For someone recovering from psychiatric hospitalization, that loss of connection can make an already difficult transition even harder.
I know people who lost their lives to suicide. I can't know what would have changed their stories, but I know how much it can mean to feel included, to laugh with friends, and to be reminded that people still want you around.
We built Still Here to help more of those moments happen.
What it does — Bringing People Back Together
Still Here is an AI-powered social reconnection agent that helps people recovering from psychiatric hospitalization reconnect with their friends and family.
When someone is struggling, even reaching out to a friend can feel overwhelming. Still Here takes on the work of remembering what brings people together, reaching out to trusted friends, and coordinating opportunities for meaningful time together.
It works in five stages:
Build a trusted circle. As part of the transition home from psychiatric care, a patient scans a QR code, has a short conversation with Still Here, and chooses which trusted friends and family members to invite.
Understand their relationships. Still Here talks with participating friends, learns about their shared interests, remembers meaningful details, and understands how they prefer to stay connected.
Recognize reconnection opportunities. Using interaction history that participants have agreed to share, Still Here identifies when recorded social connections have become less frequent than usual, without making assumptions about anyone's mental state.
Take the initiative. Rather than asking someone who's already struggling to organize an outing, Still Here reaches out to their friends, suggests activities based on shared memories, checks availability, and coordinates everyone's schedules.
Bring people together. Once the friends agree on a plan, Still Here invites the recovering person, confirms their approval, and helps everyone follow through.
An Example
Josh has recently returned home from psychiatric care.
During a conversation, his friends mention that they used to play billiards together. Still Here remembers.
Three weeks later, their recorded interactions have become less frequent.
Still Here recalls their shared love of billiards and reaches out to Josh's friends.
Maya can't do Saturday. Daniel needs to be home by seven. Alex suggests grabbing dinner afterward.
Still Here works around everyone's schedules, finds a suitable time, and calls Josh with an invitation.
Josh doesn't have to start a conversation, chase replies, or organize the outing.
He just has to decide whether he wants to go.
And for the first time in weeks, Josh is out laughing with his friends, doing something he loves. Instead of spending another afternoon alone, he's surrounded by people who are genuinely happy to see him. He's included. He's laughing. He's reminded that he matters to the people around him.
Maybe tomorrow will still be difficult. Maybe reconnecting won't always be easy.
But moments like these matter. And Still Here helps make them happen.
How we built it
We built Still Here as a voice-first AI agent that remembers relationships, detects opportunities for reconnection, and takes real-world action to bring people together.
Our biggest engineering challenge was enabling an AI agent to autonomously coordinate multiple people while protecting privacy, respecting consent, and ensuring everyone agrees before a plan is confirmed.
We combined Cartesia, Anthropic Claude, SpacetimeDB, FastAPI, React, and Google/Microsoft Calendar integrations to make this possible.
Cartesia: Real-Time Voice Conversations
We integrated Cartesia to enable natural phone conversations. Still Here can proactively call trusted contacts, understand their availability, negotiate plans, and coordinate activities without requiring patients to initiate conversations themselves.
Anthropic Claude: Agentic Reasoning
Claude powers the agent's conversational intelligence. It interprets responses, retrieves relevant memories, suggests personalized activities, and translates natural conversations into structured actions.
Rather than simply generating responses, Claude helps coordinate plans across multiple people while backend logic controls authorization and execution.
Relationship Memory and Connection Detection
We built a persistent memory system that remembers meaningful relationship details, including shared interests, past activities, and personal preferences.
A connection-pattern detector analyzes interaction history to identify when communication becomes less frequent than usual, creating opportunities for proactive outreach.
Memories are source-tracked and privacy-filtered so the agent only shares information appropriate for each participant.
SpacetimeDB: Multi-Agent Coordination
We used SpacetimeDB to synchronize proposals, scheduling constraints, and participant responses through transactional reducers.
Plans are versioned, meaning any counterproposal creates a new agreement that requires fresh approval.
No plan can be confirmed until the patient and every participating friend explicitly agree to the exact same activity and time.
FastAPI: Autonomous Workflow Orchestration
Our Python backend manages the entire lifecycle through a deterministic state machine:
Detect disconnection → Verify consent → Retrieve memories → Suggest activity → Call friends → Negotiate availability → Obtain patient approval → Confirm plan.
We implemented consent verification, quiet-hour restrictions, revocation handling, idempotency, and strict authorization boundaries. The AI can reason and negotiate, but cannot independently authorize calls or finalize agreements.
Calendar Integrations and Live Dashboard
Google Calendar and Microsoft Outlook integrations help identify overlapping availability and revalidate schedules before confirmation.
Our React dashboard uses server-sent events to visualize the agent's real-time decisions, calls, memories, negotiations, and approvals directly from persisted backend events.
What Makes Still Here Different
Most AI assistants wait for someone to ask for help.
Still Here takes the initiative. By combining persistent memory, proactive voice AI, and autonomous multi-person coordination, we built an agent that doesn't just recommend human connection. It helps make it happen.
Challenges we faced
One of our biggest challenges was designing an AI agent that could coordinate multiple people while respecting everyone's individual preferences and autonomy.
Each person has different schedules, communication preferences, and privacy boundaries. Information shared by one participant cannot automatically be shared with everyone else.
Another challenge was turning natural conversations into structured, persistent memories that the AI could retrieve and use meaningfully later.
We also had to coordinate multiple asynchronous phone calls while keeping the application's shared state consistent and updated in real time.
Beyond the technical challenges, we wanted to build something that could support people during an incredibly vulnerable period, while recognizing that human connection is only one part of recovery and does not replace professional mental health care.
Accomplishments we're proud of
What I'm most proud of is building something that has the potential to make a meaningful difference in the lives of people who are too often overlooked.
People recovering from psychiatric hospitalization face an incredibly vulnerable transition. Even when they have friends and family who genuinely care about them, depression and social withdrawal can make it difficult to maintain those relationships.
With Still Here, we wanted to do something about that.
I'm proud that we brought this idea to life by combining AI-powered conversations, relationship memory, and autonomous coordination into a system designed to help people reconnect with those who matter most.
But what excites me most isn't the technology itself. It's imagining someone like Josh, who might otherwise spend another weekend alone, laughing with his friends again because Still Here helped make that moment possible.
If what we've built can help even one person feel less alone, more connected, and reminded that they matter to the people around them, that would mean more to me than any technical accomplishment.
What we learned
Building Still Here taught us that some of the most meaningful applications of AI aren't simply about saving time or automating tasks. They're about helping people experience more of what makes life meaningful.
We learned how AI can remember the details of relationships, recognize opportunities to reconnect, and take initiative to bring people together.
On the technical side, we learned how challenging it is to coordinate multiple conversations while maintaining persistent memory, respecting permissions, and synchronizing everyone's responses in real time.
But our biggest takeaway was that Still Here's success shouldn't be measured by how many calls the AI makes. It should be measured by how many meaningful human connections it helps create.
What's next
Our next big step is bringing Still Here into hospitals and psychiatric care facilities.
We envision partnering with hospitals, mental health professionals, and psychiatric discharge programs to make Still Here available as part of the transition from hospital care back to everyday life.
Imagine a patient preparing to leave psychiatric care. Before going home, they're offered the opportunity to scan a QR code, meet Still Here, and begin building a trusted circle of friends and family who can help them stay connected during recovery.
We want to work with healthcare professionals and people with lived experience to understand how Still Here can responsibly support social connection during this critical period, alongside existing clinical follow-up and support systems.
Our next steps include strengthening privacy and consent protections, improving voice conversations and calendar coordination, and conducting real-world pilot studies to evaluate whether Still Here helps people maintain meaningful relationships after discharge.
Our long-term vision is to make Still Here accessible through hospitals and mental health programs, so that more people recovering from psychiatric care have opportunities to stay connected with those who love them.
Because sometimes, a few hours laughing with friends, feeling included, and being reminded that you matter can mean more than words can express. We want to help more of those moments happen.
Still Here
Helping people reconnect with the people they love, even when reaching out feels impossible.
Built With
- ai-agents
- anthropic
- api
- cartesia
- claude
- generative-ai
- google-calendar
- llm
- mental-health
- natural-language-processing
- outlook-calendar
- real-time
- spacetimedb
- voice-ai
Log in or sign up for Devpost to join the conversation.