Known issues first. Everything else goes to WordBlock Labs.

The biggest complaint in AI roleplay, and the reason this exists. Anti-H 2.0 and the app's guards cut it a lot but do not remove it. If a companion keeps writing your side, tell it with +ooc, and if it keeps happening across a whole session, restart the model server.

Local models often miss or mismatch quotation marks. Anti-H 2.0 helps; the app also cleans up some formatting, and we are checking whether that cleanup removes quotation marks in some cases.

Anti-H 2.0 tends to write fewer paragraphs than you ask for at higher paragraph levels. A known weak spot on the fix list.

Put the rule in Hard Rules and say it plainly, in the first or second person. Models follow specific instructions far better than vague ones. See Tell it exactly what you want.

Start your model server, load the model, then type /connect (or /connect 8080 for your port).

The first release is unsigned. The release notes will explain what you will see, with a file hash and a scan link so you can check the download.

Use the form on wordblocklabs.com and choose the SoloRoleplayer type. You choose what to send: a redacted diagnostic and optional screenshots. We never ask for your chats or saves.