Skip to content

Concepts

How Vukem Video works

One command core, many thin adapters, and a render that shows exactly what the export produces.

Last updated:

The idea#

All editing logic lives in one place, the command core, written in Go and free of input and output. A project is a document; an edit is a list of operations that the core validates and applies as one step.

Around the core sits one service. The MCP server, the CLI, the HTTP API and the browser editor are thin adapters over it. They parse a request, call one service method and format the answer. No rule about timelines is written in an adapter, so a person and an agent get the same result and the same history.

What happens on an edit#

  1. The adapter sends a batch of operations and the revision it was planned against.
  2. The core checks every operation. If one fails, nothing changes.
  3. The batch is saved as one revision with its actor, such as local:claude-code or local:ui.
  4. The result lists what was created and changed, and what the lint rules found.

Pixels#

The preview and the export use one render plan. Pixels are composed in Go; FFmpeg runs only as a separate process to decode media and to encode H.264. An agent that asks for a frame sees what the export will contain.

Nothing leaves your machine#

There is no outbound network connection by default and no telemetry. Text that comes from a media file is treated as data, never as an instruction.

Next steps#