Let your AI agent create and edit video

Video has been one of the last things an AI agent could not really do on its own. The tools assume a person: a canvas to drag, a timeline to scrub, a preview to eyeball. decodx removes that assumption.

Every decodx command is non-interactive, and every artifact it emits is schema-validated JSON. That is exactly the surface an agent needs — it can act, then read back a structured result and decide what to do next.

The loop an agent runs

• Author. From a brief, the agent writes a scene spec — narration, captures, an avatar, music.
• Render. It submits the job to the cloud and polls a typed status until the video is ready.
• Revise. To edit, it changes the spec — a headline, a clip, a duration — and re-renders. Caching means only the changed scene is rebuilt.

decodx create "60s feature tour of the new inbox"
decodx render --approve
decodx edit scene outro --headline "Try it free"
decodx render        # only 'outro' rebuilds

Costs are estimated up front and gated behind approval, so an autonomous agent proposes work rather than quietly spending. You get an agent that can genuinely own a video — from first draft to every future edit — while you stay in control of what it spends.