How Mizar's Memory Works

Most AI chatbots have the memory of a goldfish. Every conversation starts from scratch. You repeat yourself. You re-explain context. It feels like talking to someone who can never quite remember who you are.

We built Mizar's memory system to fix that.

The three layers of memory

Mizar uses three distinct layers of memory, each serving a different purpose:

1. Conversation context

This is short-term memory within a single conversation. When you're going back and forth with Mizar about a topic, it keeps track of everything said so far. Ask a follow-up question, and it knows exactly what you're referring to.

2. Personal knowledge

This is what makes Mizar feel truly personal. Over time, Mizar learns things about you:

• •Your name, location, and timezone
• •Food preferences and dietary restrictions
• •Work schedule and habits
• •Favorite topics and interests
• •People you mention frequently

This information is explicitly extracted from your conversations and stored separately. It's not hidden in some opaque model — you can view, edit, and delete any of it from your dashboard.

3. Knowledge base

This is the information you actively teach Mizar. Upload documents, share URLs, or simply tell it things you want it to remember. This creates a personal knowledge base that Mizar can search whenever you ask a question.

For example, upload your company's employee handbook, and Mizar can answer questions about PTO policies. Share a recipe collection, and it can suggest dinner ideas based on what you have in the fridge.

How retrieval works

When you send a message, Mizar doesn't search through all your memories sequentially. That would be slow and imprecise.

Instead, we use a technique called retrieval-augmented generation (RAG). Your message is converted into a semantic embedding — a mathematical representation of its meaning. This embedding is compared against all stored memories and documents to find the most relevant ones.

The top matches are injected into the conversation context before the AI generates a response. This means Mizar's answers are grounded in your actual data, not hallucinated.

Privacy by design

A few important things about how we handle memory:

You own your data. Everything Mizar remembers is visible in your dashboard. You can view it, edit it, or delete it at any time.

We don't train on it. Your memories are never used to train or fine-tune AI models. They're used exclusively to improve your personal experience.

It's encrypted. All stored memories are encrypted at rest and in transit.

You can wipe everything. One click and all your data is permanently deleted. No backups, no retention period.

The result

After a few conversations, Mizar starts to feel less like a generic AI and more like an assistant that actually knows you. It greets you by name. It remembers your coffee order. It knows that when you say "the kids," you mean Emma and Lucas.

That's the experience we're building toward — AI that's genuinely personal.