Video editing,
automated.

We built a custom editing workflow around an AI agent powered by Codex with Astra and HyperFrames. It assembles recordings, B-roll and approved graphics into videos. Reference films and successive revisions shape its editing style. Approved styles, rules and corrections are saved in project files and the Behave library, building a production memory the agent uses on subsequent videos. The Behave team previews and approves the resulting material.

90%: of the video editing process automated
60 h: saved every month

Scope: AI video editing agent
Context: Behave content production

Let's talk about your workflow

Behave Studio interface visualisation for creating an episode, choosing a format, and adding source materials and editing instructions. — New episode · interface visualisation. View full size (opens in a new tab)

Behave Studio interface visualisation showing video productions, an Akita episode preview and editing feedback. — Interface visualisation of the deployed system. View full size (opens in a new tab)

Technologies

Codex + Astra

The agent's brain

Understanding instructions, planning and controlling tools.
HyperFrames

Scenes, animations and rendering.

Technical foundation

Python
Automation and quality control.
JavaScript / Node.js
Tools and composition logic.
HTML / CSS / GSAP
Scenes, captions and animations.
FFmpeg / FFprobe
Media processing, encoding and checks.
Whisper
Local transcription of recordings.
JSON / Markdown
Project state, approved styles, rules and corrections.

Video editing,automated.

Technologies

Technical foundation

Video editing,
automated.