Focuspia

Focuspia turns focus from a lonely timer into a quiet shared place: friends create a room, choose a to-do, and keep each other company without a camera or call.

Comment

Inspiration

I run a focus timer service. For a long time, I had wanted to build a version that combines focus timers with gamification and social features. The idea itself is not new, but building a real service with personal spaces, realtime interaction, and social features requires significant development effort. While working on other projects, I kept putting it off.

I had used Codex 5.5 as my main development tool for several months. When Codex 5.6 arrived, I felt that I could delegate a more ambitious project and trust it with more context. OpenAI Build Week appeared at the right time.

Focuspia began as an experiment: could I start from an empty repository and build and deploy a working full-stack product in roughly 20–30 hours using a single Codex project session?

What it does

Focuspia is a 3D social focus space.

Users can sign in with email, create and decorate a personal room, add friends, and visit each other’s rooms. They can move around the room and send short speech bubbles in realtime.

Each todo supports either an elapsed-time stopwatch or a countdown timer. When a user is focusing, their focus state and elapsed time appear above their avatar. The app also includes a public cafe space for a more structured shared-focus environment.

Room decoration has separate local-editing and published states, so visitors only see the version that the room owner has chosen to share.

How we built it

Focuspia started from an empty repository.

The client uses React, Vite, Electron, React Three Fiber, Three.js, Tailwind CSS, and Supabase Auth. It renders licensed low-poly GLB assets and uses a grid-based room-decoration system.

The realtime backend uses Phoenix Channels, Phoenix Presence, and PostgreSQL. Phoenix handles rooms, friendships, focus sessions, published room layouts, and realtime state. Supabase is used for email authentication and token verification.

Codex 5.6 was the primary implementer for this project, not just a code-completion tool. I specified the intended technology stack and assets, then provided high-level product and UX guidance. I reviewed the results and gave feedback in natural language, but I did not write any application code myself.

Codex created the codebase from the empty repository, divided broad requests into related work, implemented features, investigated bugs, ran builds and tests, and supported deployment. During the initial build, it worked on the Phoenix backend and the Electron/Vite client in parallel. The same workflow continued through the 3D room, realtime interactions, focus tracking, UI iteration, and web deployment.

Challenges we ran into

Time was the main constraint. The project started from an empty repository, but required a 3D client, authentication, a realtime backend, persistence, and deployment.

The harder challenge was delegating broadly without losing the product direction. I needed to review the actual results, identify what felt wrong, and give feedback in product language instead of specifying every implementation detail in advance.

Connecting authenticated web clients to Phoenix realtime channels and keeping local room edits separate from published room state also required repeated testing and iteration.

Accomplishments that we're proud of

In roughly 20–30 hours, Focuspia went from an empty repository to a deployed full-stack app.

The result includes email authentication, a 3D room, furniture placement, friend relationships, room visits, realtime avatars, speech bubbles, focus timers, focus status, persisted focus records, and a public cafe scene.

I am also proud that I could guide and review a full product build without writing application code myself, while Codex handled the implementation from start to finish.

What we learned

This project changed how I think about AI-assisted development.

I am more comfortable working from product intuition, concrete feedback, and an evolving sense of what feels right than from a fully specified plan written upfront. Codex 5.6 worked well with that process: I could provide the surrounding context, react to what it built, and describe problems in natural language.

I do not see this as a replacement for judgment or planning. I still provided the initial direction and decided what was ready to ship. But Codex felt like a development partner that could turn broad feedback into appropriately scoped work and help an intuitive builder move quickly without first translating every idea into a detailed engineering specification.

What's next for Focuspia

Focuspia began as a hackathon experiment. I wanted to see how far I could take a full-stack product with Codex 5.6 in a short amount of time. The result was more satisfying than I expected, and I now see enough potential to continue developing it beyond the event.

I plan to develop Focuspia into a real commercial service. The next steps are refining the social focus experience, improving the cafe and room interactions, adding proper avatar models and animation, connecting focus tokens to furniture purchases, and making the web experience easier to join and use with friends.

I also plan to keep building in this way: providing product direction and feedback while using Codex as the primary implementation partner.

Built With

Share this project:

Updates