Chat

A box is a directory of your data, kept under version control with a full history of changes (using git); a card is a markdown file with a structured header; the agent is the coding agent (Claude Code or Codex) that operates the box. Chat is the everyday way you talk to it.

What it does for you

What it needs

Nothing beyond the box itself and a working agent login (Claude Code or Codex). See ../install/index.md.

How it works, briefly

Every conversation is a durable chat card, so it survives reloads and is browsable like anything else in the box. The agent acts inside a turn and asks you a question when it is unsure rather than guessing; its mid-turn work (files touched, background notes) is visible in the transcript. Chat runs on demand, when you send a message, not on a schedule. The box screen sends a thought to whichever chat it fits (quick chat); see web-interface.md. A box with no working agent login shows a notice in place of the message box and sends the owner to the Admin page to connect one.

Limits

The not-delivered mark is kept for at most 24 hours, and a crash (as opposed to a restart) loses a queued message without leaving the mark. A chat that cannot be resumed (for example, an old session format) says so rather than silently starting a new, disconnected one. The documentation does not describe a message-editing feature; a sent message stands as sent. Chat search matches the words as typed and does not understand meaning, by choice, so that your transcripts are not sent to an outside service to build it; it covers only chats on the machine the box runs on, and only the text you and the agent wrote, not tool output or attachments. Searching chats from the box-wide search is not built.

Go deeper

../reference/cards/chat.md, ../reference/cards/chat-thread.md (history only), ../reference/chat-voice.md, ../reference/narration-mode.md, voice.md