tired of telling ChatGPT who Donna is every conversation, so I built the version with memory that actually keys on people

Enhancing AI Memory: Building a Personalized, Person-Centric Chat App

In the realm of AI assistants like ChatGPT and Claude, a common limitation has been their handling of memory. Typically, these models treat user data as a flat collection of facts, lacking the ability to dynamically structure and relate information about different individuals in your life. This often results in repetitive explanations—such as having to reiterate who Donna is every time you want to discuss her—a frustrating experience for users seeking more fluid and context-aware interactions.

Recognizing this gap, I spearheaded the development of a specialized chat application designed to address this exact challenge: implementing structured, per-person memory that evolves organically from your ongoing conversations. Rather than merely storing isolated facts, this system intelligently extracts and organizes information about each individual you mention, creating a personal, evolving profile for friends, colleagues, or family members.

How It Works

When you vent about someone like Donna—sharing anecdotes or observations—the app captures these insights as time-stamped facts linked directly to her profile. Each entry includes a reference back to the original message, allowing for traceability and context. Over weeks and months, this accumulates into a comprehensive, nuanced picture of the people who matter to you, written entirely in your own words.

Key Features

• Structured Per-Person Memory: Unlike generic memory models, this design maintains dedicated profiles for each individual, enabling richer and more relevant contextual references during conversations.

• Automated Fact Extraction: The app runs asynchronous processing on a lightweight model to analyze your messages and update each person’s profile seamlessly without adding latency to the chat experience.

• Privacy-Conscious Architecture: All message content remains confidential. Analytic processes exclude message data, utilizing secure scrubbing methods, and privacy-preserving data handling practices are implemented throughout the system.

• Flexible Provider Support: The app supports multiple AI providers, including Claude Sonnet 4.6 and GPT-5.1, with easy toggling per conversation, ensuring adaptability to diverse needs and preferences.

Experience It Yourself

Interested in seeing how this personalized memory system works? You can try a demo without creating an account here: Demo Link. Simply start chatting, vent about someone in your life, and then explore the ‘People’ section to see how your ongoing conversations have built a detailed profile.

This project exemplifies how targeted enhancements in AI memory management can significantly improve the utility and personalization of virtual assistants. By focusing on structured, dynamic profiles for the people that matter, users can experience more natural, efficient, and meaningful interactions with AI.

---

Note: This development is ongoing, and feedback from users like you helps shape the future of personalized AI companions.