Project Concept

AmbientMind is a voice-native “second brain” that runs in the background and continuously captures what you say. Instead of relying on memory or manual note‑taking, it listens to your conversations, transcribes them in real time, extracts key insights, and lets you ask follow‑up questions that are enriched with live web search.

Our goal is to make ambient capture actually useful: not just raw transcripts, but structured understanding (key points, action items, decisions) plus an agent you can talk to about what just happened. AmbientMind combines ElevenLabs speech‑to‑text and text‑to‑speech, Google Gemini 2.5 Flash for reasoning and summarization, our custom Search API for up‑to‑date information, and a Neon + Drizzle backend for session persistence.

What We’re Working On

• Refining the ambient listening loop (latency, deduplication, noise handling)
• Smarter summarization and question‑answering over recent sessions
• Better UI for browsing sessions, transcripts, and sources
• Tighter integration between search results and conversational context

How Others Can Contribute

• Voice / UX: Design better voice and chat flows, prompts, and interaction patterns
• AI / Agents: Improve summarization, search‑routing, and tool‑calling logic
• Infra / Data: Enhance the session data model, add analytics, and improve reliability
• Frontend / Design: Polish the Tangerine theme, animations, and session browser UX

Tech Stack

The project is built with Next.js 16, React 19, Bun, shadcn/ui, Tailwind CSS v4, Drizzle ORM, and integrates Google Gemini and ElevenLabs. It’s designed to be hackable: contributors can easily extend the agent’s tools, add new “skills” (e.g., calendar integration or task syncing), or plug in additional model providers.