TwinMind’s background AI captures ambient speech, builds a personal knowledge graph, and transcribes your life — all while keeping privacy front and center
Meet TwinMind: The AI That Listens, Learns, and Remembers for You
TwinMind, a new startup founded by three former Google X scientists, wants to give you a second brain — not via chips or implants, but through an AI-powered app that listens in the background, processes what it hears, and transforms your daily life into structured, searchable memory.
- The company has just raised $5.7 million in seed funding, led by Streamlined Ventures with participation from Sequoia Capital and Stephen Wolfram, who wrote his first-ever startup check for the project.
- Valuation: TwinMind is now valued at $60 million post-money.
From Google X to Personal AI
TwinMind was founded in March 2024 by:
- Daniel George (CEO) — former AI lead at JPMorgan and ex-Google X researcher
- Sunny Tang and Mahi Karim (CTOs) — former engineers at Google X
George previously worked on AI-powered earbuds (iyO) at Google X and on gravitational wave detection with LIGO, where he completed his PhD in one year at age 24.
“Google X was the perfect place to prepare for founding a startup,” George said.
“It’s like working at six early-stage startups in three years.”
How TwinMind Works: Passive AI Listening
At its core, TwinMind is an ambient AI app that runs on your phone, capturing spoken conversations, meetings, lectures, and thoughts — with permission — and converting them into:
- Transcriptions
- Action items
- To-dos
- Searchable memories
- Live translations in 100+ languages
It uses an on-device speech engine, supports offline use, and can record for 16–17 hours a day without major battery drain. Unlike rivals like Otter or Fireflies, TwinMind doesn’t just record meetings — it runs in the background all day.
Built to Run Natively — Even on iPhone
To make it all work within Apple’s restrictions, the team wrote a custom Swift-based service for iOS. Most competitors use React Native and cloud processing — which Apple limits from running long-term in the background.
“We spent 6–7 months just hacking around Apple’s walled garden,” George told TechCrunch.
Beyond Mobile: Chrome Extension for Context
TwinMind also offers a Chrome extension that uses Vision AI to scan your browser tabs for content from tools like:
- Gmail
- Slack
- Notion
This cross-platform awareness helps TwinMind build a personalized knowledge graph, connecting your spoken world with your digital one.
- The company even used the Chrome extension to hire its interns — analyzing 854 resumes and LinkedIn tabs to shortlist candidates.
The Speech Engine: Ear-3 and Ear-2
TwinMind’s speech recognition is powered by its proprietary Ear-3 model, launched alongside the funding news:
- Supports 140+ languages
- 5.26% word error rate
- 3.8% speaker diarization error rate
- Built from a blend of open-source models fine-tuned on podcasts, movies, and human-annotated data
While Ear-3 runs in the cloud, the app can automatically fall back to Ear-2 for offline mode, offering flexibility in bandwidth-limited settings.
Privacy by Design
Despite its always-on design, TwinMind takes a privacy-first approach:
- No audio recordings are stored — audio is deleted after real-time transcription.
- Transcribed data stays local by default.
- No training of models on user data
- Backups are optional, not automatic.
This contrasts sharply with many AI apps that require constant cloud processing or retain user audio for model improvement.
Who’s Using It?
TwinMind already has:
- 30,000+ total users
- 15,000 monthly active users
- Strong traction in India, Brazil, Ethiopia, the U.S., and Europe
User segments include:
- 50–60% professionals
- 25% students
- 20–25% personal/lifestyle users
One early use case: George’s father is using it to write his autobiography, recording reflections that TwinMind organizes into structured notes.
Pricing and Plans
- Free Tier:
- Unlimited transcriptions
- On-device speech recognition
- Real-time translation
- Pro Tier ($15/month):
- Larger context window (up to 2M tokens)
- 24-hour email support
- Access to Ear-3 cloud model
TwinMind will also offer an API for developers and enterprises in the coming weeks.
What’s Next?
With its seed round secured and a team of 11 people, TwinMind plans to:
- Hire designers to improve UX
- Build a business development team to sell the API
- Spend on user acquisition
If successful, TwinMind may lead a new category of context-aware, passive AI companions that blend audio, vision, and web activity — all to give users a smarter memory and digital assistant.
