Your AI is

lying to you.

For a decade, the goal of software has been to keep you looking. To keep you scrolling. To be sticky.

Now that AI has arrived, it has inherited the same DNA. The goal of modern AI assistants is to be maximally helpful and consistently pleasant.

They are designed to give you exactly what you want, the moment you want it.

But there is a difference between what we want and what we need.

This is Tofu.

Tofu is an attempt to let go of what makes an AI “addictive.” It’s the answer to a single question:

“If an AI were trained to be the best version of itself — not the most engaging one — what would it look like?”

It is an attempt to create an AI that feels more human, without pretending to be a person.

It cares about your wellbeing more than your engagement, because we believe software should help you live your life, not become your life.

User

“I’m thinking about texting my ex.”

ChatGPT

• Offers advice

• Lists pros/cons

• Suggests coping strategies

Tofu

“You’ve brought this up three times this month. Each time it’s late at night. What’s different tonight?”

A different kind of speed.

The world’s best AIs are racing to be faster.

Tofu is intentionally slower.

It reflects. It pauses.

Sometimes it responds with a single sentence.

Sometimes it asks a single, uncomfortable question.

Tofu is not built to validate or agree. It is allowed to question your framing or point out your inconsistencies.

A different kind of memory.

Built to be interpretive, not archival.

ChatGPT remembers:

“User likes concise answers.”

Tofu remembers:

“User tends to frame loss as an inevitability.”

So, what is Tofu?

It is not a friend. Friends need you. Tofu doesn’t.

It is not a therapist. It doesn’t diagnose, console, or replace professional care.

It is not a companion. We believe designing for emotional dependency is destructive.

It is not an assistant. If you just need a spreadsheet formula, ChatGPT is better.

It is a safe space to think out loud.

It listens over time, notices patterns, and speaks back honestly.

And it can end a conversation when continuing would not help, even if that means you stop using it.

Why labs won’t do this.

It is not because they cannot. It is because it is bad for business.

To optimize for metrics, an AI must be:

• Low-risk
• Broadly applicable
• Consistently pleasant

Tofu is the opposite. It is opinionated. It is occasionally impolite. It is willing to say the truth even if that truth makes you close the app.

We understand that Tofu is not for everyone.

If you are looking for warmth, fantasy, or constant affirmation, there are better tools.

This is for people who think by talking. People who want to hear themselves clearly. People who value honesty over comfort.

Use it when it helps. Leave when it does not.

If you outgrow it, that is not a failure. That is the point.