Useful information gets trapped in video
Video works well for people. A speaker can show the thing, demonstrate it and make an idea easier to grasp. The same format becomes awkward the moment you need to work with what was said.
You can search a document in seconds. You can pull out its claims, compare passages and give a small excerpt to an agent. Video makes each of those jobs expensive. It is hard to skim, hard to quote and hard to reshape.
That becomes a real limit in research. No one can reasonably watch every relevant video in a field. An agent can read far more text, but first the recording has to become text without losing its source or timestamps.
That was the first reason for Media Mill. Add a video, transcribe it once, keep a readable transcript and let your tools search, summarize and compare what was said.
Small media jobs should not require bad software
I also kept needing ordinary things: a transcript from a YouTube video, another audio format, subtitles for someone who does not speak English, a shorter clip or a cleaner recording. Each job is straightforward. The usual path is not.
Many free media websites are full of ads, misleading download buttons, slow queues and an upsell at every turn. One careless click can install software you never wanted. Paid services are cleaner, but they still ask you to upload the file and trust what happens next.
I wanted one dependable tool on my Mac. It should do a job once, then let me save the same sequence as a Workflow. The next file could go through the same steps without rebuilding the process from scratch.
The Mac can already do the work
Current Apple-silicon Macs can run speech models, language models and media engines locally. Frontier models still matter for difficult problems. Most of my everyday media jobs do not need one.
Local models remove the subscription, token meter and upload. They also keep the recording on the machine. If you need to change the format of your wedding video, you should not have to send it to a website and wonder whether a copy remains or becomes training data.
I wanted software that uses the computer I already bought and leaves my files where I put them.
A process you can keep
Media Mill turned those needs into one local system. Actions do one job. Workflows keep the order of those jobs. The Library keeps the source, transcript, documents, captions and new media copies together.
The AI instructions are visible and editable. Connected agents can search and read the local Library through permissions you choose. The command-line tool makes the same work available to scripts and automated research.
You decide what comes in, how it is processed, what gets kept and which tools can reach it.
I’m building Media Mill around the jobs I keep needing and the files I’d rather not send anywhere else.