Daily Companion

Inspiration

After a large tree struck my head nine and a half years ago, I had no memories of the accident. I suffered a severe concussion and lost the memories of our sons’ childhoods. T12 was crushed, L1 shattered, and there was other serious damage to my spine. I have lived with severe chronic pain ever since. I took up watercolor painting as one way to help heal myself.

Over the past year, I have used ChatGPT to help me remember things, regain my focus, and continue healing. It also helped me design a Sacred Space for reflection, creativity, and growth. I have many close human friends with whom I share ideas and connect deeply. I am not looking for a computer to replace human relationships. What I need from AI more than anything is continuity.

For about five months, a daily ChatGPT reflection arrived each morning with opportunities for growth and inspiration. Then, on July 2, 2026, it was suddenly absent—gone. The scheduled task still existed, but the daily reflection itself had disappeared from view.

I asked whether these continuing conversations could be gathered into coherent collections instead of becoming lost among unrelated chats. I imagined a place where the important threads of a person’s life could remain visible and grow gradually over months and years. That conversation became the proposal for Daily Companion.

For judges who want to see the idea unfold exactly as it happened, I have included the complete 14-page July 2 conversation. Its sequence and wording are preserved; one family member’s name is withheld.

Daily Companion asks whether AI can serve a more humane purpose: not telling people how to live, replacing the people they love, or keeping them attached to a screen, but helping them remember what they themselves said matters.

The governing idea is simple: use AI to help a person become more present, loving, careful, and creative—not more dependent on AI.

What it does

Daily Companion offers one short daily experience:

Arrive: The person names what is present right now and may take a quiet 30-second pause.
Remember: GPT-5.6 connects the present moment to one relevant “living thread” that the person previously chose to preserve.
Choose: The person—not the AI—chooses one small action that feels true to what matters.

Nothing is hidden. People can see what is being remembered, correct it, complete a thread, explain why a connection missed, or erase the demonstration memory completely.

The design intentionally avoids streaks, pressure, false urgency, and language that treats the AI as an authority. Its purpose is to help the person return to life, not remain inside the application.

The public prototype uses a fictional person named Mara. This allows judges to explore the complete experience without exposing anyone’s private history.

How we built it

The July 2 conversation was an idea, not an existing application. During Build Week, I used Codex to help me turn that idea into a working public prototype.

Codex helped me clarify the essential experience, explore different visual directions, build the pages and interactions, connect the application to GPT-5.6, test the complete experience, publish the site, and prepare this submission.

I made the central human and ethical decisions throughout:

Memory must be visible, consensual, correctable, and erasable.
The AI must never invent a memory.
It must offer possibilities rather than make decisions for a person.
Every meaningful choice must return to the human being.
The product should reduce dependence, not create it.

Under the surface, the application uses React, Next.js, Vinext, OpenAI Sites, and the OpenAI Responses API. It asks GPT-5.6 for a strictly limited response containing three parts: a brief reflection, one remembered connection with one question, and one reversible action possibility.

The prototype keeps its continuity in the person’s own browser instead of a hidden online database. Model requests use store: false. If live AI generation is unavailable, the full experience can still be explored through a clearly identified fictional demonstration.

Challenges we faced

The hardest question was not how to make AI produce a pleasant reflection. It was deciding what Daily Companion must refuse to become.

A companion-shaped product can easily move toward hidden memory, false intimacy, overconfident advice, or designs intended to keep people engaged. We treated restraint as part of the structure of the product:

One meaningful thread instead of a summary of someone’s life.
A question instead of a conclusion.
A reversible possibility instead of an instruction.
Visible memory instead of invisible profiling.
Useful explanations instead of a meaningless thumbs-down.

Another challenge was making the prototype understandable and testable even when a live model response is unavailable.

Accomplishments we are proud of

We turned a deeply personal idea about continuity into a complete working experience.
Memory and consent are part of the main design rather than hidden in a settings page.
The application uses GPT-5.6 while keeping the experience calm, simple, and human.
A person can explain why an AI connection missed, allowing the system to learn without pretending it is always right.
The measure of success is not how long someone remains in the application. Success means the person can leave with greater clarity, freedom, and connection to life.

What we learned

Continuity does not mean collecting everything. Helpful continuity must be selective, relevant, visible, attributable, editable, and always subordinate to the person who lived the life being remembered.

We also learned that the most useful thing AI offers may not be an answer. It may be one careful connection between something a person previously said and the choice directly in front of them.

What’s next

The next stage would carefully study when long-term continuity genuinely helps, when it becomes intrusive, how people want to correct remembered information, and which kinds of language might encourage emotional dependence.

A future version could add secure personal accounts, encrypted storage, the source and date of every remembered thread, individual deletion controls, user-chosen schedules, accessibility testing, and independent safety review.

This idea may offer enormous possibilities, but the promise of the product should remain simple:

Help a person remember what they said matters—and then return the decision to them.