Kailash Labs

Kailash Labs describes itself as a video research and product lab. Its stated mission is to convert the world's videos into structured data that machines can reason over, using small, grounded video understanding models that can be adapted to a downstream video analysis task with little data. The lab's thesis, laid out on its home page, is that general video models now transfer across detection, retrieval, captioning and temporal grounding, that efficiency matters more than raw capability because camera feeds run at 30 to 60 frames per second, and that video is the missing context layer for AI agents. It lists robotics, security, industrial monitoring, sports and media as the fields it wants to work with, and it names Qwen3 VL as the base model for a model called Marlin. The product is an SDK. The vendor says it collapses the work of choosing a model, sampling frames, wiring a camera and evaluating on your own footage into a few lines of code, with the ability to swap models, connect live feeds and run evals so an agent can see video. Early SDK access is currently open to design partners, who are invited to get in touch through the site. The vendor does not publish pricing, and there is no public signup or download. Access is by contacting the team.