Writing with a model

The model writes in your files, in a mode you choose per piece of work. What it earns its place doing differs between a book, a paper and a policy.

The shape of a turn

Talk the chapter through, then let it draft into the file. Your uncommitted edits are snapshotted first, so the turn starts from a known point and you can always get back to where you were. The turn is then committed with the model that produced it named in the message.

That is why the mode matters more than the prompt: the record shows which one was in use, so "the model did the reading" and "the model wrote the argument" do not end up looking the same afterwards.

Hosted or local, per work

Byline talks to hosted models and to models running on your own hardware, and the choice is per work rather than global. Which one wrote a given turn is in the commit, so switching provider partway does not blur the record. Tokens and cost are accounted per work as you go.

What it will not do

A write that cannot be attributed is refused rather than saved. That is the one place the application is deliberately inconvenient: there is no path that puts text in the manuscript without the record knowing where it came from.

Single author per repository today. Review comes back through DOCX rather than through shared accounts.