# Giizu

> Giizu is the video layer for AI agents: infrastructure where agents see, edit and ship video, right where the footage lives.

Video is the largest kind of data on the internet, and the one AI agents still can't work with where it lives. Today an agent must download footage before it can understand or edit it, and looking at one hour of 720p video at one frame per second costs about 4.4 million tokens.

## How Giizu works

- Upload once: footage stays in Giizu's cloud.
- Understood on arrival: transcripts, scene descriptions, faces, shots, contact sheets and a 720p proxy, organised in layers from a roughly 10,000-token project brief down to single frames.
- Agents read instead of watching: Claude, Codex or a custom agent connects over MCP or the API and reads those layers.
- Agents edit with their own code, next to the footage: FFmpeg, OpenCV, segmentation models and OpenColorIO are preinstalled.
- Preview streams in seconds, with no export or download.
- Finish on the originals: the same code re-runs on the camera files for the final render.

## Primitives

- read: zoom from the project brief down to a single frame
- run: execute the agent's own code beside the footage
- preview: stream every change back in seconds
- render: re-run the same code on the camera originals

## Status

Giizu is onboarding a small group of early teams building with video and agents.

## Links

- Website: https://giizu.com
- Contact: gauransh@giizu.com
- GitHub: https://github.com/giizu-hq