Memoir

💡 Inspiration

I started building Memoir out of deep personal frustration. I was exhausted from trying to manage my own life across five different apps—notes, calendars, reminders, and to-do lists. It felt like I was working for my productivity tools instead of them working for me.

When I looked at existing AI chatbots, I realized they were incredibly smart but lacked persistent personal memory; they forgot who I was as soon as the chat ended. On the other hand, traditional productivity apps had memory but absolutely zero intelligence. We have tools that remember but have no intelligence, and tools that have intelligence but no memory.

I realized there was a massive gap for a truly private, intelligent companion that remembers context and takes proactive action. I wanted to build an assistant that always remembers—giving your memory a giant upgrade so you never forget a thing.

🧠 What it does

Memoir is a private, AI-powered personal execution system. Instead of opening multiple apps and typing out your life, you simply talk to Memoir. It acts as your mind's auto-save.

Memoir processes your natural conversation to extract and store persistent facts (e.g., "my landlord is Mr. Ade" or "I prefer morning workouts"). Over time, it builds a highly personalized memory graph. It leverages this deep context to execute tasks, set proactive reminders, and intelligently plan your schedule. It doesn't just respond to queries; it turns your passing thoughts into structured actions.

⚙️ How we built it

Memoir is built from the ground up to be flawless, prioritizing a calm interface and local-first privacy.

• Frontend & Architecture: The app is built in Flutter using a strict MVVM (Model-View-ViewModel) architecture. Views handle pure UI, while Riverpod handles state management and business logic.
• Local-First Privacy: To ensure personal data remains completely private, we use Hive for blazing-fast local storage. Facts, timers, and tasks never leave the device unless explicitly required.
• The AI Engine: Instead of relying on a standard, off-the-shelf LLM chat loop, we engineered a custom orchestration pipeline. This pipeline leverages multi-model reasoning (via Groq, Gemini, and OpenRouter) to break down user intents into actionable jobs. It manages context dynamically, ensuring the AI can access your memory graph and execute system actions without overwhelming the context window.

🚧 Challenges we ran into

• Token Bloat & Context Limits: Feeding long chat histories into an LLM quickly causes context overflows and slows down response times. We overcame this by designing a custom routing architecture that handles intents systematically rather than recursively, drastically reducing the token load.
• AI Hallucinations on System Actions: Building a reliable execution system on top of non-deterministic AI models is incredibly difficult. We had to implement strict, custom validation layers between the AI's output and the app's internal logic to guarantee that timers, reminders, and memory updates execute flawlessly every time.
• Architectural Discipline: Maintaining a clean UI while handling asynchronous AI streams, background tasks, and local database syncs required heavy reliance on Riverpod to keep the frontend completely decoupled from the heavy business logic.

🎓 What we learned

Building Memoir was a masterclass in AI orchestration. I learned that the secret to a great AI assistant isn't just having the smartest foundation model, but rather how you orchestrate the data around the model. I learned how to move beyond basic chatbot wrappers to build deterministic, action-oriented AI pipelines, while also deepening my expertise in Flutter and local persistence strategies.

🚀 What's next for Memoir

Memoir is just beginning its journey to become a daily operating system for personal life. Next steps include:

• Expanding the internal registry to support deeper integrations with calendars and journaling.
• Enhancing proactive background capabilities that anticipate user needs without prompting.
• Securing funding to move Memoir from its foundational phase into the market, giving people their time back and reducing daily cognitive load.

Built With

• dart
• envied
• flutter
• gemini-api
• go-router
• groq
• hive
• llama-3.1
• openrouter-api
• riverpod