Meta Journal

Inspiration

We realized that for AI to be truly useful, it needs more than just a chat window—it needs a train of thought. We wanted to build a tool that doesn't just bookmark links but captures the mental context of why you saved something, effectively giving both the user and their AI a persistent, organized memory.

What it does

Meta Journal is a local-first browser extension that turns your scattered browsing history into a cohesive, searchable tapestry of knowledge.

Smart Snips: Highlight and capture text from any webpage. Meta Journal anchors it to the source, preserving the exact scroll position and visual context. AI Contextualization: Using Anthropic, Meta Journal automatically generates titles, summaries, and tags for every thread, understanding the content instantly. Bi-directional Pinning: The "magic" feature. Click a thread in Meta Journal, and the browser "teleports" you back to the exact URL and visual coordinate where you had the thought, even if you closed the tab days ago. Contextual Chat: Chat with your captured context. Ask questions across completely different websites and let Meta Journal find the connections. Voice briefing: Briefs a users research

How we built it

The Chrome Extension: Built with React, TypeScript, and TailwindCSS (Manifest V3). It handles the DOM manipulation for "snifting" content and manages the side panel interface. The Backend: A robust Python FastAPI server that manages data persistence. The Brain: We used ChromaDB for vector storage to enable semantic search across preserved context. The Intelligence: AnthropicAPI powers the summarization, auto-tagging, and conversational capabilities, turning raw text into structured insights.

Challenges we ran into

Bi-directional Scrolling: Getting the browser to accurately scroll back to a specific text snippet on a dynamic web page (which might have changed or reflowed) was incredibly difficult. We had to implement fuzzy matching and robust DOM selectors to ensure the "teleportation" feature felt magical and reliable. Context Window Management: Balancing the amount of context effectively passed to the AI without overwhelming the token limit or causing latency was a constant tuning process. Chrome Extension Limitations: Working within the sandboxed environment of Manifest V3 while maintaining persistent connections to a local backend required creative architectural solutions.

Accomplishments that we're proud of

The "Teleport" Effect: We successfully implemented the deep-linking scroll feature. Seeing the browser instantly snap back to a specific paragraph from days ago feels genuinely futuristic. Seamless AI Integration: The latency between "Snip" and "Summary" is almost invisible. It feels like the AI is reading with you in real-time. Premium Aesthetic: We didn't just build a tool; we built a product. We're proud of the design language—clean, dark-mode first, and distraction-free.

What we learned

Context is King: Raw data is useless to an AI without metadata. By preserving the source coordinates and metadata, the AI's utility grew exponentially. Local-First is Hard but Worth It: Managing state locally instead of relying entirely on the cloud reduced latency significantly and improved the privacy posture of the app. Prompt Engineering is UI: We learned that the quality of the UI is directly tied to the quality of the system prompts governing the summarization and tagging.

What's next for Meta Journal

Multi-Modal Snips: capturing images and charts, not just text. Obsidian/Notion Sync: Automatically exporting threads into a user's "Second Brain" notes app.

Built With

• anthropic
• elevenlabs
• react
• tailwindcss
• yutori