Conversations that resolve.

A thread holds the context for a single question and resolves to a single output. That constraint is the magic. The model doesn’t blur because there are 200 messages from twelve unrelated topics in the window — it sees the thread, scoped to the channel, with the artifacts and profiles the channel says to load. The same model, with sharper input, produces sharper output.

When you start a thread, Boardbox asks (implicitly): what will this resolve to? An artifact update, a Linear ticket, a Notion doc, a calendar event. The thread tracks its target. When the answer is good enough, you close it. Internal artifacts update with version history; external actions still require approval before they push.

When a thread resolves, a small model runs over its messages and writes a short summary. The summary lives in the channel index — the table of contents your agents read every turn. The next thread in the same channel can reference the work that came before without you re-explaining and without the agent re-reading 200 messages.

If a thread is the kind of work you’ll repeat — morning standup, weekly investor brief, daily ad report — hit Make recurring. Boardbox extracts a brief from the thread, proposes a schedule and a tool set, and you land in the agent review screen. The conversation you just had becomes the agent you didn’t have to design from scratch.