Active Memory, Passive Forgetting: A Different Philosophy for AI
MeteorWish Team · 7 min read
Most AI assistants are built to remember.
Every conversation stored. Every question logged. Every half-formed thought sitting in a database somewhere.
What if the problem with AI isn't that it forgets too much — but that it remembers too much?
MeteorWish is a private AI assistant that doesn't store your chat history by default. It forgets by default, and only remembers what you choose to save.
Why AI's Passive Memory Is a Problem
Most AI products today use what we'd call passive memory — they remember everything by default. You don't choose what gets stored. You don't control what gets used. You just... talk, and hope for the best.
This creates a practical problem:
You start self-censoring.
Think about it. When was the last time you asked an AI something truly personal? Something you wouldn't want stored in a database somewhere?
You're not alone in this hesitation. According to Pew Research (2023), 70% of Americans have little to no trust in companies to make responsible AI decisions. And a KPMG Consumer Trust Survey (2024) found 63% of consumers are specifically concerned about generative AI compromising their privacy.
The numbers point to a structural issue: the default incentive is to keep data.
Policies vary by product, plan, and region, but many AI products may retain chats and use them to improve models unless you opt out. Some developers may keep this information for long periods. A Stanford HAI write-up examining AI developers' privacy policies highlights long data retention periods, training on sensitive data, and a general lack of transparency about what happens to your conversations.
A Deloitte Connected Consumer study (2025) of 3,500 US consumers found that 24% of gen AI users report having experienced a data-privacy issue firsthand — not just a fear, but something they believe already happened.
The result? AI becomes a tool you use carefully, not a space where you think freely.
Active Memory, Passive Forgetting: A New Approach
We built MeteorWish on the opposite principle.
Passive forgetting. Conversations are not stored on our servers by default. In normal operation, we don't log chat content or use it for training. When you start a new conversation, the previous one is gone from our systems. Not archived. Not "anonymized." Gone.
Active memory. When something matters — an insight, a decision, a piece of information you want to keep — you save it explicitly. One click. That goes to your personal memory system.
The key difference: conversations don't persist, but your chosen memories do.
You control what crosses the line from ephemeral to permanent. Not the platform. Not an algorithm. You.
How We Ensure Deletion
This isn't just a policy promise — it's architectural. Within MeteorWish, our system is designed so that chat content never hits persistent storage in the first place. There's no database of chats to delete because there's no database of chats, period.
Your saved memories are stored encrypted in the cloud and remain under your control. Everything else simply doesn't exist after the session ends.
(We'll publish a technical deep-dive on our deletion architecture soon.)
Does MeteorWish Store Chat History?
No. MeteorWish doesn't store chat transcripts or conversation history by default.
The only content we keep is what you explicitly choose to save as a memory.
We do keep minimal usage metadata (like timestamps, token counts, and which model was used) for billing and reliability — but this metadata never includes your chat content. For details, see our Privacy Policy.
When Ephemeral AI Matters
Career crossroads. You're considering leaving a stable job for something completely different. Maybe it sounds crazy. Maybe you're not sure yourself. You need to think it through — explore the fears, weigh the tradeoffs, articulate what you actually want. Save the insight that crystallizes. Let the messy process disappear.
Creative exploration. You have a half-formed idea. It might be brilliant. It might be ridiculous. Probably both. You need a space to riff, to push it further, to find out what's actually there. No permanent record of the dead ends. Keep the gems that emerge.
Personal processing. Some things you just need to talk through — relationship tensions, family dynamics, inner conflicts you're trying to understand. The value is in the conversation itself. You don't need a transcript. You need the clarity that comes from thinking out loud.
These are the moments where thinking out loud matters most. But they're also the moments where people hold back with mainstream AI assistants. Because everything is being recorded.
AI should be a good thinking partner — available, patient, and low-stakes. But when the default is retention, people naturally hold back.
How MeteorWish Works: Ephemeral AI in Practice
MeteorWish is a full AI assistant. We route requests to leading AI providers and pick the right model for the job, rather than locking you into a single one. You can:
- Have deep conversations — Work through problems, brainstorm ideas, get feedback on your thinking
- Get things done — Research, writing assistance, analysis, the usual AI capabilities
- Build a personal knowledge base — Your saved memories become a searchable, organized collection of insights that matter to you
What about providers?
A reasonable question is what happens once a request leaves MeteorWish. We route requests through an AI gateway and configure providers for privacy-by-default (for example, opting out of training on customer content where available).
The core idea stays the same: MeteorWish doesn't keep a transcript of your conversations, so your history doesn't accumulate with us over time.
The difference isn't what the AI can do. It's what happens to your conversations afterward.
Many mainstream AI assistants store conversations by default. Some offer deletion controls, but by then the data may have already been processed (and in some cases used for training) and can linger in backups.
MeteorWish stores no chat content by default. There's nothing to delete because nothing was stored. And because we route your questions to the best available model — not a single provider — you get premium AI capabilities without turning your conversation history into a dataset.
This is what we mean by "ephemeral AI": an assistant where chat content is temporary by default, and only what you explicitly save becomes long-term memory.
Why We Charge Directly
MeteorWish is $20/month. Your first month is $12.
Here's the thing about free products: when you're not paying, you're usually the product. Your data becomes the business model. That creates a fundamental tension between your privacy and the company's incentives.
We charge directly, so your data stays yours.
No ads. No training on your conversations. No selling access to your profile. Our only incentive is to make the product good enough that you want to keep paying for it.
No free trial, but your first month is just $12. That's our way of letting you try with low commitment.
A Different Relationship with AI
Most AI products market memory as a feature: remember more, personalize more.
We think there's another way.
An AI that respects boundaries. That forgets unless asked to remember. That treats your conversations as yours to control.
Not an all-knowing oracle. More like a thoughtful companion that's fully present, without carrying baggage from the last conversation.
Some will say this limits the AI's usefulness: "How can it help you if it doesn't know your history?" We think the tradeoff is worth it. The memories you explicitly save are higher-signal than a comprehensive log you never asked for.