# Chat modes & autonomy

Canonical URL: https://docs.perisclaw.com/chat-modes

You are always in control of how active Perisclaw is. There are two layers: one
**overall setting** that decides how forward it may be in general, and
**per-chat exceptions** for the chats you want handled differently.

Your **self-chat** sits outside these modes. It is your private workspace and
always works, unless you block it with WhatsApp’s built-in Chat Lock.

## The overall setting

Set this once on the **Permissions** page. It answers a single question: when
may your assistant act in a chat that is not your self-chat?

| Mode        | When it acts outside your self-chat                                                                                             |
| ----------- | ------------------------------------------------------------------------------------------------------------------------------- |
| **Strict**  | Only when you summon it with `/claw`.                                                                                           |
| **Default** | Also in chats you explicitly hand over, for the purpose and period you specify. This is where most people should sit.           |
| **Auto**    | Also on new inbound activity in eligible one-to-one chats, following the [engagement rules you write](/permissions-and-access). |

For a standing rule such as *“when a sales lead messages, notify me”*, create
an [Automation](/automations). For a finite outcome such as *“coordinate a time
with Mira and book it”*, create a [Mission](/missions). Both stay narrowly
scoped and do not turn a whole chat into an always-on assistant conversation.
You don't have to pick; ask for what you want and Perisclaw sets up the right
one.

## Per-chat exceptions

Whatever your overall setting, individual chats can be handled differently.
Tell your assistant, from inside the chat or from your self-chat:

* *“Participate in this thread until Friday and answer as my assistant.”* It
  may now respond visibly there, for that purpose and period.
* *“Stop participating in this chat.”* Back to only acting when summoned.
* *“Never read this chat.”* Blocked entirely. See below.

Both lists live on the **Permissions** page: **Participating now** shows every
chat where your assistant currently has explicit permission to respond, with
how to stop it; **Blocked chats** shows the chats it must leave alone, which
apply in every mode, including Strict.

## Blocking a chat properly

**Warning: Only blocking is a privacy control**

  There is a real difference between{' '}
  *“don’t bother me about this chat”* and{' '}
  *“never read this chat.”* Say
  ** “never read \<chat>”** in your self-chat and your
  assistant cannot read, search, summarize, quote, or reply in that chat at
  all. If you later ask it to do something in a blocked chat, it tells you the
  chat is blocked and asks whether to unblock, rather than quietly doing it.

## Draft-first by default

Whenever Perisclaw is about to send a message to someone else on your behalf,
it shows the draft in your self-chat and waits for your OK. Approve by replying
*“send”* or reacting 👍 to the draft.

The exceptions are `/claw` used directly in a chat (you are already standing
in it, so it acts there) and chats you have handed over. The first time it
messages a new contact for you, it confirms the name **and** the number before
anything goes out.

## Whose voice it speaks in

When Perisclaw sends something for you in another chat, it goes out as **you**,
with no assistant header. Nobody needs to know how the message got written.
When you summon it with `/claw`, it answers as the assistant, clearly labelled.

If you would rather it be openly the assistant in a particular chat (a work
group where people should know they are talking to your EA), say so: *“act as my
assistant in this group,”* and every message there is labelled until you flip
it back.

For what your assistant can reach in the first place, see
[Permissions & access](/permissions-and-access).

### [What it can reach](/permissions-and-access)

The access and approval rules behind every mode.

### [Learn the everyday basics](/use-cases)

Self-chat, /claw, and how drafts and approvals work.