A video series that runs itself, after you approve it

PageToVid already drew up a plan for a website: a list of video ideas, each with a brief and the shots it needs. Until now you acted on it one idea at a time.

Now an assistant can fetch the whole plan, including which ideas have already become a video, and make the missing ones. Called without confirmation, it returns a free preview of exactly what would be made. With confirmation it renders them, each under its own idempotency key per idea, shape and language, so a retry never makes the same video twice.

The planning engine now lives in one shared library used by the website, the VS Code extension and the MCP tools. Three doors, one implementation, so they always agree on what “already made” means.

A show is a template plus a feed. The template is a storyboard that stays the same every episode; the feed supplies new content each time. A feed can be RSS, Atom, JSON or a simple manual list, and the reader is tolerant going in and strict going out: every item leaves with a stable key, plain text and full addresses, whatever the source looked like.

Episodes are ordinary videos. They are made by exactly the same steps as any filmed video or animation, so editing, re-rendering and embedding all work on an episode without knowing it belongs to a show.

You create and run shows from an assistant or VS Code, with four tools to create, list, run and update them. The site’s page has a Shows panel where you give the first approval and can pause or resume a series at any time.

Automation that spends on a schedule needs a stronger rule than “be careful”. Ours is this: a show may never spend on its own before someone has seen what it makes. In practice:

1. 1a new show starts on hold, and the scheduler never runs a show on hold
2. 2running it without confirmation gives a free preview of the episodes it would make
3. 3running it with confirmation makes the episodes and approves the show, which from then on may follow its schedule
4. 4every run has a cap on episodes, and every day has a cap on spending
5. 5a feed item can never become two episodes, guaranteed by a unique key and an idempotency key together

If the balance runs low, a run stops cleanly between episodes, never halfway through one. A manual list cannot have a schedule at all, because there would be nothing new for it to pick up.

That shape, preview first, an explicit approval, then caps, is now our rule for every automation that spends. It comes from a simple observation: the person who sets up a schedule is not watching when it fires. If the first paid run were also the first time they saw the result, a template with one wrong line could turn into a week of wrong videos.

A first approval costs one extra click. In exchange, every recurring series on PageToVid is something its owner has seen at least once, and the caps keep the worst case small if a feed changes shape later.

The same release made every MCP tool say up front what it does with your balance: free, reading your account, or spending. A test fails if a tool that spends is missing from the list of tools that charge.

This first release brings the engine, the tools, the endpoints for the editor and the Shows panel on the site. Three things are next on the list: a dedicated shows view inside the VS Code extension, checks that compare episodes across a whole series so it stays consistent from week to week, and switching on the hourly scheduler. Until it is on, a show runs when you press run.

To drive shows from your editor, start with the VS Code extension, or connect an assistant from the assistants page.