Remember your thoughts before they're forgotten.

discover page
companion
thoughts
saved
companion talking
thought detail
matching

Inspiration

Most thoughts never become anything.

They show up while we are driving, standing in the shower, or lying in bed half awake, and disappear almost as quickly as they arrived. Even when we write them down, they often sit forgotten inside a notes app. Social media does not feel right either, because unfinished and personal thoughts are not always meant to become content.

We built LOT, Language of Thoughts, to create something in between.

LOT is a private-first place where people can write what is actually on their minds, return to those thoughts later, and gradually understand how their thinking changes over time.

The idea behind LOT is simple:

Notes remember what you wrote. LOT tries to remember what you have been thinking.

What it does

Users can write thoughts privately and decide later whether they want to keep them private, share them with friends, or publish them publicly.

The AI Companion does more than summarize an individual entry. With the user's permission, it can notice recurring questions, tensions, emotional patterns, and changes across multiple thoughts over time.

For example, instead of only responding to one note, Companion may eventually notice:

a question the user keeps returning to
an old thought that has become relevant again
a belief that appears to be changing
language that has become more confident or less apologetic
a topic that has quietly disappeared from the user's writing

LOT also supports safer sharing. When someone chooses to publish a private thought, the original remains unchanged. LOT prepares a separate public version that helps remove names, contact information, usernames, links, and identifying details while trying to preserve the meaning and emotional tone. The user reviews the result before anything is published.

Public thoughts can then be discovered through meaning and resonance rather than only popularity. Users can explore similar thoughts and, where appropriate, connect with people who seem to be thinking around the same question.

How we built it

LOT includes an iOS application, a web application, and a Firebase backend.

The iOS app was built with:

Swift
SwiftUI
SwiftData
WidgetKit

The web experience was built with:

TypeScript
HTML and CSS
esbuild

The backend uses:

Firebase Authentication
Cloud Firestore
Firebase Cloud Functions
Firebase Cloud Messaging
Firebase Crashlytics
Node.js

We used OpenAI models as part of LOT's reflection and understanding layer. The system analyzes opted-in thoughts to produce structured signals such as themes, tensions, safe public rewrites, recurring patterns, and potential connections between thoughts.

Rather than sending every interaction through the largest model, we designed the AI as a layered system. Routine processing can happen through smaller or cheaper paths, while higher-level reasoning is reserved for tasks where judgment matters more, such as identifying meaningful changes over time or deciding which thoughts may be worth resurfacing.

We also built a longitudinal data foundation that records compact monthly signals without storing raw note text in the analytics layer. These signals can later support evidence-based observations such as:

“You have written about this less often over the last three months.”

or:

“The way you describe this relationship appears to be changing.”

The goal is to make each observation traceable to real user history rather than generating vague, horoscope-like statements.

Challenges we faced

Making AI feel useful without replacing the user's thinking

It is easy to build an AI that summarizes text or gives advice. It is much harder to build one that helps a person notice something themselves.

We repeatedly redesigned Companion to avoid generic validation, excessive advice, and chatbot-style responses. The strongest experience was not when the AI sounded intelligent. It was when the user understood their own thinking more clearly.

That led to one of our main design principles:

Companion should not make users feel that AI understands them. It should help users understand themselves.

Building long-term memory without becoming invasive

LOT needs memory to become valuable over time, but private thoughts are among the most sensitive forms of user data.

We therefore designed memory around explicit consent, encrypted private notes, compact structured signals, and separation between private originals and public versions. Raw private text is not placed into the longitudinal counters used for future pattern comparisons.

Preserving meaning during safer sharing

Removing a name or phone number is easy. Removing identity risk while preserving the actual thought is much harder.

A poorly designed privacy filter can turn an honest thought into broken or meaningless text. We built the public-sharing flow as a separate review process so the private original remains untouched and the user always stays in control.

Creating a calm product instead of another engagement machine

Many social products are built around counts, followers, streaks, and constant notifications. Those mechanics would conflict with LOT's purpose.

We worked to make thoughts, rather than popularity, the center of the interface. Discovery prioritizes relevance, specificity, and quality. Companion resurfaces meaningful moments rather than manufacturing daily activity. Public sharing is intentional rather than automatic.

What we learned

We learned that the hardest part of building an AI reflection product is not generating more responses. It is knowing when the system has something worth saying.

Sometimes the most honest Companion experience is silence.

A weak insight shown every day quickly becomes noise. A rare observation supported by months of actual writing can become unforgettable.

We also learned that private writing and public conversation do not need to be separate products. A thought can begin privately, become clearer through reflection, and only later become something the user chooses to share.

That journey, from private thought to understanding to connection, became LOT's core loop.