Rumi Labs

Rumi Labs is an AI media company that transforms passive video watching into interactive experiences. Its core business is a browser extension and desktop app that runs a vision-language model locally on users' devices, indexing what they watch without recording or storing video. Users earn points for valid viewing contributions, which help train the company's cultural intelligence model.

The company's key advantage is its proprietary VLM architecture, which outperforms Gemini 2.5 Pro on narrative comprehension tasks while being 100x smaller. It uses a "Narrative Intelligence" Mixture-of-Experts approach to recognize relationships, causality, and emotional context in visual media. Rumi also operates a decentralized network that enables compliant access to copyrighted media content that centralized AI labs cannot reach.

The market pain point is that millions of hours of IP-protected media remain untapped by AI due to copyright constraints, despite people spending ~4 hours daily watching content. Rumi solves this by running AI processing on users' devices, respecting IP while enabling contextual advertising, interactive features, and deep viewership analytics.

In the past six months (March–September 2026), Rumi has been building out its desktop application and expanding its distributed VLM network, allowing everyday computers to participate in media indexing. The company has also released several audio-related models on Hugging Face, including text-to-speech and audio-text-to-text models, suggesting expansion beyond visual media understanding.