stackmemory.app

StackMemory is a "Second Brain" for your code. Powered by Gemini 3.0, it indexes your entire repo to let you chat with your project, debug visually, and onboarding teammates in seconds.

Comment

💡 Inspiration

I have been developing for over 1.5 years now, and after building many projects, I bumped into a common issue: How I documented my projects left me wondering. I had notes, links, and disparate docs, but no comprehensive way to document and save everything. I tried tools like Obsidian and others—all great, but not exactly what I wanted for code.

That is the basis of StackMemory. It bridges the gap between manual documentation and automatic context, ensuring that no project is ever "forgotten" again.

🧠 What it does

StackMemory is an intelligent, AI-powered dashboard that acts as a Second Brain for your code.

  • Automatic Ingestion: A CLI tool (stackmem sync) that silently scans your local projects and syncs their metadata, dependencies, and structure to the cloud.
  • GitHub Integration: Seamlessly connects to your repositories to fetch, index, and understand your code structure automatically.
  • Vibe Coder (AI Assistant): A multimodal chat interface powered by Google Gemini 3.0 Flash that allows you to "talk" to your codebase. It uses RAG (Retrieval-Augmented Generation) to answer complex architectural questions.
  • Stack Intelligence: Automatically outlines your tech stack, detects security vulnerabilities, and suggests upgrades.
  • MCP Bridge: Implements the Model Context Protocol to connect cloud AI with your local environment (Postgres, Filesystem) securely.
  • Service Locker: A unified vault for all your critical project links (AWS, Vercel, designs).

🏗️ How we built it

We built StackMemory using a modern, edge-ready stack:

  • Frontend: Built with Next.js 16 (App Router) and React 19 for high performance and Server Actions. We used TailwindCSS v4 and Framer Motion for a premium, clean UI.
  • Backend & Data: Supabase handles our PostgreSQL database, Authentication, and Vector Storage (pgvector) for our semantic search engine.
  • AI Engine: We leverage Google Gemini 3.0 Flash via the Vercel AI SDK for its massive context window and speed, essential for processing large codebases in real-time.
  • Infrastructure: The app is deployed on Vercel, utilizing Edge Functions for low-latency responses.

🚧 Challenges we ran into

  • Context Window Limits: Even with modern LLMs, feeding an entire codebase into a prompt is expensive and slow. We had to implement a smart RAG (Retrieval-Augmented Generation) pipeline that chunks code semantically and retrieves only the most relevant snippets for the AI.
  • The "Silent" CLI: Building a CLI that works seamlessly across different environments (macOS, Linux, Windows) and syncs data without interrupting the user's workflow was a significant engineering hurdle.
  • MCP Integration: Implementing the nascent Model Context Protocol to bridge the gap between our web-based AI and the user's local filesystem required deep dives into experimental specifications.

🏅 Accomplishments that we're proud of

  • The "Vibe Coder": Creating an AI that genuinely "understands" the code context rather than just guessing. It feels like pair programming with the original author of the code.
  • Zero-Config Onboarding: The fact that a user can run stackmem sync and have a fully populated dashboard in seconds is a major UX win.
  • Aesthetics: We didn't just build a tool; we built a place developers want to be in. The neumorphic/dark-mode UI is crafted with care.

📚 What we learned

  • Metadata is King: The more structured data (versions, dependencies, file types) you can feed an AI, the better its reasoning becomes. Raw code isn't enough; context is everything.
  • Developer Experience (DX) is fragile: One extra step in a CLI tool can drop adoption by 50%. "Silent" automation is the only way to ensure tools are actually used.

Built With

  • framer-motion
  • google-gemini
  • lucide-react
  • next.js
  • postcss
  • react
  • supabase
  • tailwindcss
  • typescript
Share this project:

Updates