Keep your own recordings

The problem on the rest of this site is that a company holds your history and can change what it charges for it. One answer is to hold the history yourself. Boswell is free software that records from your Omi onto your own computer, transcribes it, works out who was speaking, and lets you search all of it. This page is honest about what that costs you, because it is not free of effort.

Your Omi keeps working

The device you already bought connects to it. You do not need different hardware, and you do not need to give the device up.

The recordings land on your computer

Audio, written versions, and the names of who spoke are files on a disk you own. Nothing is stored on someone else's server, so nobody can raise the price of getting to it.

Nothing renews

There is no account and no plan. If you stop using it, nothing expires and nothing is taken away.

You can read the whole thing

It is open source. Anyone can check what it does with the audio, which is a claim you can verify instead of a promise you have to accept.

WHAT IT ACTUALLY DOES

The same job, on your side of the wire.

It records continuously, writes down what was said, tells voices apart and remembers them, so a person you name once is recognised in later conversations. It groups clips into conversations, searches every line by word or by meaning, and can have a model read a conversation and write down the facts and tasks in it. That model can run on your computer too.

Read the code and the setup notes ↗

WHAT IT IS NOT

It is not a replacement product.

There is no phone app, no cloud account, and nobody to email when it breaks. It is software that a person wrote for their own recorder and then wrote down carefully enough that other people can run it. Judge it as that. If you want something that works the moment it arrives, this is not that.

WHAT IT COSTS YOU

None of this is free of effort.

Holding your own history means running the thing that holds it. Here is what that actually asks of you, so you can decide before you spend an evening on it rather than after.

It runs on a computer you leave on

A desktop or a small always-on machine at home. Not a phone. A laptop that sleeps will miss recordings while it is asleep.

Transcribing needs a graphics card, or a paid key

With an NVIDIA graphics card it keeps up easily. Without one it can send the audio to OpenAI or Deepgram instead, which costs money per hour of audio.

With neither, it falls behind

On an ordinary processor it works out about two and a half times slower than the recordings arrive. That is fine for catching up overnight. It is not fine for wearing all day.

You set it up yourself

There is no installer and no support desk. It is a project, not a product. If that sounds like a bad afternoon, this is not for you yet.