AI That Remembers Your Life: How Two-Tier Memory Works

Chat with any popular AI assistant — ChatGPT, Gemini, Claude — and close the app. Open it again tomorrow and say "Remember what I told you about my Goa trip?" Blank stare. It has no idea what you're talking about.

This is the fundamental flaw of most AI assistants: they have amnesia. Every conversation starts from zero. You're a stranger every time.

VivaLyn Sathi works differently. It uses a two-tier memory system that lets it remember your name, your family members, your food preferences, your medication schedule, and hundreds of other personal details — across sessions, across days, across months.

Why memory matters in a personal AI

A personal AI without memory is just a chatbot with a fancy name. The whole point of a "companion" is that it knows you. It should know you prefer chai over coffee, that your mother's name is Sunita, that you're allergic to peanuts, that you have a meeting every Tuesday at 10 AM, and that you've been stressed about the appraisal.

Without memory, every interaction requires re-explaining context. With memory, the AI builds on what it already knows — making each conversation more relevant, more helpful, and more personal.

Tier 1: Short-term memory (conversation buffer)

Short-term memory holds the current conversation thread. When you're talking to Sathi, the last 20 messages stay in context. This allows natural, multi-turn conversations:

• "Remind me to call the dentist."
• "Actually, make it next week instead."
• "And add a note that it's for the root canal follow-up."

Sathi understands "it" refers to the dentist reminder and "next week" modifies the original date — because the previous messages are still in the buffer. This is standard for any modern LLM chat interface.

The key difference: when the conversation ends, instead of discarding everything, Sathi runs an extraction step. Important facts are pulled out and saved to long-term memory.

Tier 2: Long-term memory (persistent store)

Long-term memory is where the real magic happens. After each conversation, Sathi extracts key facts and stores them in a local database on your phone:

• Personal facts: Your name, age, occupation, city, family members
• Preferences: Favourite foods, allergies, sleep schedule, exercise habits
• Routines: Morning routine, work schedule, weekly patterns
• People: Who you mention frequently — their names, relationships, and context
• Events: Important dates — birthdays, anniversaries, medical appointments
• Health: Medications, conditions, doctor names, prescription details

When you start a new conversation, Sathi loads relevant long-term memories as context. So when you say "What time is my meeting tomorrow?" — it already knows your typical Tuesday meeting schedule. When you say "Add paneer to the grocery list" — it knows you buy paneer every week because it's part of your routine.

Memory extraction: how facts are pulled from conversations

Not everything you say gets stored in long-term memory. Sathi uses a summarisation model to identify what matters:

• Named entities: People, places, dates, amounts
• Expressed preferences: "I don't like spicy food" → stored as a preference
• Repeated patterns: If you mention gym 3 times in a week, it recognises a routine
• Explicit requests: "Remember that I'm vegetarian" → stored immediately

Casual filler ("hmm", "okay", "nice") and transient details ("the traffic was bad today") are not stored — unless they contain a named entity or expressed emotion worth tracking.

Conversation history: find anything you ever said

Beyond the two-tier memory system, Sathi stores complete conversation transcripts locally. Every message is timestamped and full-text searchable. You can ask "What did I tell you about the Goa trip?" and Sathi retrieves the exact conversation — even if it was weeks ago.

This is different from memory. Memory stores extracted facts. History stores the full conversation. Together, they give Sathi both wisdom (what it learned about you) and recall (what was actually said).

Privacy: all memory stays on your phone

This is the critical difference from cloud-based AI memory (like ChatGPT's memory feature). With Sathi, all memory — short-term, long-term, and conversation history — is stored in a local SQLite database on your phone. Nothing is uploaded to a server. Nothing is synced to the cloud.

The server processes your message and immediately forgets. It's the phone that remembers. This means your personal facts, preferences, health details, and life context never leave your device.

Delete the app, and all memory is permanently erased. There's no server-side backup to worry about.

The result: AI that feels personal

After a week of talking to Sathi, it knows your morning routine, your favourite foods, your family members' names, and your stress triggers. After a month, it anticipates your needs. After three months, it feels like an AI that actually knows your life — because it does.

That's the difference between a chatbot and a companion. Memory.