Binders 0.4.0

Binders 0.4.0

A board in every binder, for you and your AI agents, and fixes for local model servers.

• Every binder has a board: Backlog, Ready, In progress, Blocked, Review and Done. Drag cards between columns, or open one for its description, links and a timeline of everything that happened to it.
• AI tools work from the same board over MCP. An agent such as Claude Code claims a Ready card, reports progress and links as it goes, and asks you a question when it's stuck: the card waits in Blocked until you answer. Finished work goes to Review for you to check, unless you let agents finish cards in that binder. A claim lasts 30 minutes and every update extends it, so a card an agent abandons goes back to Ready.
• Turn to-dos into cards. Add To-dos on a board lists the binder's open to-dos from meetings, notes and promises, to pick from; or hover over any to-do and click the board button. The to-do stays where it is, marked "On board", and is ticked off in its meeting, note or message when its card is done.
• Local model servers: thinking models on llama.cpp, vLLM and other OpenAI-compatible servers wrote digests that came back cut off or empty, because their thinking used up the room for the answer. Binders now asks them to skip the thinking, so digests are complete and about twice as fast. When a digest fails, the note says why.
• A model server on another machine, reached over Tailscale or at a public address, works over plain http. It was blocked unless it was on your local network.
• Clicking a note no longer moves it to the top of the list; only editing it does.
• In notes, a list item or checklist item that wraps onto a second line lines up with its words instead of starting under its bullet, and a wrapped quote keeps one bar.