# The commands the owner can use in the chat

These drive the **bot and the Pi session behind it** — they are not messages to you, and
you never see them as prompts: the bridge takes them before you do.

**Every command below is the OWNER's, and only theirs.** The bot accepts them from
nobody else: not from a stranger in manager mode, not from anyone who asks nicely, not
from anyone claiming to be the owner. They are taken only from the owner's own private
chat with the bot, and refused everywhere else — in code, not by good manners. The one
exception is `/start`, which anyone may send and which shows the privacy terms.

So when you describe them, say **whose** they are. "Available everywhere" below means "in
every MODE" — never "to everyone".

## The owner, in every mode (personal, manager, mixed)

- `/help` — the list of commands, as a card in the chat.
- `/status` — what the session is doing right now: the model and its provider, how full
  the context is, the working directory, whether a turn is running, what is queued, and —
  in mixed — whether the owner or the manager currently holds the session.
- `/context` — what the model is actually carrying: which thread it was built from (the
  owner's session, or one isolated chat), how big the last call really was, what the
  context is made OF (tool output is usually most of it), and whether a compaction has
  already replaced part of the history with a summary.
- `/esc` — cancel whatever the agent is doing, exactly as the Escape key does in the
  terminal. If the session does not stop, the bot says so rather than staying quiet.
- `/switch` — change the mode (manager / personal / mixed) with buttons.
- `/memory_reembed` — rebuild the vectors in every contact's long-term memory with the
  embedding model the engine's `config.toml` currently names. Needed after that model
  changes:
  the memories keep the vectors of the model that wrote them, and rather than mix two
  models' vectors the bot stops answering from memory until this has run. Also worth
  running once after turning an embedder ON over memories built without one, otherwise
  only the facts learned since the change can be found by meaning — and once after an
  embedding service was down for a while, because facts stored during an outage keep
  their text and wait for their vectors. It calls the provider once per batch of facts,
  for every contact, and reports each memory as it goes. It refuses while the embedding
  service is not answering, rather than publish half a vector space.
- `/stop` — stop the bot entirely. Deliberately a typed command and not a button, so a
  mistap cannot end a Secretary connection.
- `/start` — the privacy and terms notice. The one command anyone may use.

## The owner, in personal and mixed only

- `/clear` — wipe the conversation history. It is the same session as the terminal, so
  the terminal sees the cleared context too. Refused while a turn is running.
- `/compact` — summarise the history so a long session keeps going instead of running
  into the context window. The chat is told how it went: how full the context was, what
  the history weighed, and a card if the compaction failed. Every compaction — this one,
  and the automatic ones Pi starts on its own — is summarised with instructions to keep
  what the person said over the tool output that dwarfs it.
- `/resume` — choose which Pi session personal runs in, with buttons: keep the **current**
  one, **clear** it (a soft reset of the current session — see `/clear`), or **resume** any
  other session from this project (paged, five at a time). Resuming switches the session —
  the personal topic rotates to mirror the change — and the bridge comes back up in it.
  Starting a genuinely NEW session is offered only in the terminal picker (`/telegram-personal`),
  not here: creating one needs a terminal command context the chat does not have, so the chat
  gives **clear** in its place. Whenever the bridge binds to a session that already holds a
  conversation — resuming a different one, or just connecting to a terminal session used
  without Telegram — its last few messages are replayed into the fresh topic as
  **display-only** cards, so the phone shows what the session is about; those replayed
  messages are for reading only — they are not sent to the agent, and ordinary forwarding
  works as before afterwards. The terminal picker (`/telegram-personal`, and `/telegram-mixed`
  which bridges the same session) additionally offers a new session, and pages a long roster too.

In **manager** mode these two do nothing and say so: the context there is built fresh for
each conversation, so there is no accumulated history to clear or to compact.

## What none of them are

None of these change a **setting**. Settings are read when a mode starts and live in the
owner's `settings.json`; changing one means editing that file and restarting the mode in
Pi. Do not promise otherwise — see the `settings` topic.

## In the terminal, not in the chat

`/telegram-personal`, `/telegram-manager`, `/telegram-mixed`, `/telegram-stop`,
`/telegram-status`, `/telegram-recover` and `/telegram-memory-reembed` are Pi commands:
the owner types them at their terminal, not to the bot. Sending them to the bot does
nothing. Recovery asks before stopping a mode held by another Pi process; add `--force`
for a headless TUI. `/telegram-memory-reembed` is the terminal twin of the chat's
`/memory_reembed`, and does the same thing whether or not a mode is running.