Bots and sections

A bot has a name, face, model, soul, skills, and memory. A section is one persistent conversation with that bot. Start separate sections for separate tasks; each has its own context and history.

Give a bot a role

Use Bot settings to configure how it works:

The teammate description tells other bots when to ask this one for help. For example, “Reviews database changes and checks migration safety” is more useful than “Helpful assistant.” If you leave it blank, Hexbot derives a description from the soul.

Faces and colours

In Profile, pick a shape and a colour for the bot’s face, or use a photo. A picked face is kept with the bot on the daemon, so the web app, the desktop app, and the mobile app all draw it. The bot’s message bubbles take a light wash of its face colour in sections and rooms, so you can tell bots apart at a glance. A bot with a photo keeps grey bubbles. The mobile app shows faces and can switch a bot to a photo, but faces are picked in the web or desktop app. On an older daemon a picked face is saved only as a picture: the web and desktop apps still draw it live, and the mobile app shows the picture.

Keep a conversation going

Reopen a section to continue the same task. Closing the app or restarting the daemon does not delete it. Each section keeps a persistent Pi conversation; a fresh section starts with a new context window and the bot’s current settings.

Soul, selected skills, tool definitions, and the teammate list stay fixed for an existing conversation. Start a new section after changing them so the bot receives the new setup. Credentials and runtime settings can update separately; a temporary section model override does not change the bot’s default model.

The context meter at the top right of a section shows how much of the model’s context window the conversation fills, as a percentage, with the plain numbers on hover. A tick on its bar marks the summary point: when the conversation exceeds it, Hexbot summarises the older messages so the section can go on, and the pill reads “Summarising” while that happens. If the model’s window is too small to keep any messages before that point, the tooltip says older messages are summarised after every reply. After a summary, the meter reads “Context” until the next reply is measured, even after the daemon restarts. The summary keeps the thread of the work but not every detail, and a long section makes the bot slower and less accurate, so when the meter is high, start a new section for the next task. The bot can still search earlier sections when it needs them.

Section titles

A new section takes a title from your first message. The bot can rename its own section as the topic becomes clearer, and the roster shows that title. Names you set yourself are kept until you or the bot change them.

Use a title that lets you find the task later. The conversation history is also searchable by the bot, so a new section can build on earlier work without carrying the whole transcript in its context.

When the bot searches, it leaves out the section it is in, which it already has in view. A deleted section stops showing up in the bot’s next search.

Stop, archive, and delete

Use Stop to interrupt a running answer. The section stays available, and you can send another message.

Archiving hides a section from the active list and keeps its history. To bring it back, find it in Settings, Archive, where you can search titles, first messages, and bot names, or filter by bot, and choose Restore. The bot’s Sections settings list its archived sections too. Deleting removes that section and its history permanently. It leaves the bot’s memory alone; edit Memory separately when you want to remove a remembered fact.

Deleting the bot removes its profile, sections, memory, and scheduled jobs. Stop its active work first. Back up the daemon’s data directory before removing a bot you may need again.

Work with other bots

In a room, everyone sees the shared conversation. With Message other bots enabled, a bot can instead ask one of your teammates for help in a private conversation. The bot it asks does the work itself and can’t hand the request back. Open the help row under its reply to read that exchange.

Delegate creates an isolated subagent for a subtask and returns the result to the originating conversation. It is a separate capability from asking a named teammate. Both use model tokens and the daemon’s usage limits.