Docs
Browse the docs
Docs / Senses & Autonomy

Senses & Autonomy

Senses let your sidekick keep an eye on things for you, like your inbox and the news feeds you care about, even when nobody is chatting with it. Autonomy decides how far it may go with what it notices: just keep notes, tell you, draft replies, or act.

How Senses work

Every so often your sidekick takes a look at everything it's watching. We call each look a Pulse. On each Pulse it reads anything new, decides what matters, and handles it at the level you allowed. Most of the time nothing needs doing, and that's a normal, quiet Pulse.

You'll find Senses on the sidekick's Sheet, under the Senses tab. The top of the tab tells you whether it's running, for example Watching 2 sources ยท last looked 3m ago. To make it look right away instead of waiting, press Check now.

The Senses tab, showing a news feed and a Gmail inbox being watched at the orient level, with buttons to watch a feed, connect Gmail or add another mailbox
The Senses tab. Each source has its own level, its own instructions and a Pause button.

Before you start

Senses need a Mind Module that's been tested for unattended decisions. On the Mind tab, each Mind Module shows โœ“ Senses or โœ— Senses. If your sidekick is set to Single model, or to a module marked Senses cannot run, its Pulse won't run.

If the status says Not running โ€” no schedule, check the Mind Module first. If it says Scheduled, watching nothing yet, everything is set up and it only needs a source.

Adding things to watch

Under What {name} watches you can add three kinds of source:

1

A news or blog feed

Click + Watch a feed and paste the feed's address (RSS or Atom). Then give it a name, choose its Freedom (see levels) and click Add feed.

2

A Gmail inbox

Click Connect Gmail and sign in with Google. This is the best way to connect Gmail.

3

Any other mailbox

Click + Other mailbox (IMAP), then enter the email address, an app password, the IMAP server, port and folder. The connection is tested before anything is saved, so a typo fails right there and doesn't break silently later.

Mailboxes are opened read-only. Nothing gets marked as read, flagged or deleted, and the first check reads only the 25 most recent messages. Use an app password (most providers, including iCloud, require one) so you can revoke it any time.

Instructions for a source

Each source has an Add instructions link. Use it to explain the context in plain language, for example "This is my work inbox. Anything from the accounting team is urgent."

Pausing or removing a source

Pause stops checking a source and keeps its setup. Remove deletes the source. Anything your sidekick already learned from it stays.

The four levels

Levels build on each other: each one includes everything in the levels before it.

LevelWhat it means
observeWatches and keeps a record. You won't hear from it.
orientAlso learns, and tells you what's going on.
decideAlso drafts replies and proposes work for you to review. Never sends.
actAlso does things for you, using the abilities you've switched on.

You set the sidekick's overall level on the Autonomy tab. Each source can be held lower than that level, never higher. So a noisy news feed can sit at observe while your inbox runs at decide.

No level lets your sidekick send email for you. Even at act, it can only use the abilities you've switched on. For example, it can create a task only if the Scheduled Tasks ability is on.

Rules for particular people

Under How {name} handles particular people, click + Add a rule to give standing instructions about a sender or a whole company:

  • Who: an email address, or a bare domain such as client.com to cover the whole company.
  • Call them: what your sidekick should call them.
  • How to handle them: for example "A paying client. Draft a reply and flag anything about contracts or money."

Rules change how a message is handled, never what your sidekick is allowed to do. A domain only matches that exact domain, so a lookalike such as client.com.attacker.net won't match. If your mail server can't confirm a sender really is who they claim to be, the rule is withheld and the message is treated as a possible impersonation.

What your sidekick has done

The What {name} has done section keeps a record of every Pulse, in four tabs:

TabWhat's in it
DraftsReplies waiting for you. Keep or Discard each one. Nothing is sent from here.
DecisionsWhat it decided about each item, including anything it held back. Forget undoes something it learned.
LearnedFacts it picked up, which feed its memory. Use Forget on anything that's wrong.
ChecksA log of each Pulse, so you can tell "quiet because nothing mattered" apart from "quiet because it stopped".

The Autonomy tab

The Autonomy tab on the Sheet controls how much your sidekick may do on its own, and how often it looks.

  • Autonomy: the overall level (observe, orient, decide or act).
  • Pulse: how often it looks. Choose from Every 15 minutes, Every 30 minutes (the default), Hourly, Every 4 hours, Morning and evening (8am and 8pm) or Once a day (a single morning pass). Less often is cheaper, and it handles more at once.
  • Pulse brain: the Mind Module that does the judging. Automatic uses the sidekick's own cloud thinking, so Pulses keep running whether or not your computer is on. Only modules tested for unattended decisions are listed.
  • Pulse reasoning: how hard it thinks on each Pulse. Leave it on the preset default unless you have a reason. More reasoning makes decisions steadier, but it costs more and takes longer.

Below those settings, What that permits lights up the actions allowed at the current level. If an action is allowed but the ability it needs is switched off, the page tells you so. Sources held lower lists any sources capped below the sidekick's level.

Pulses and briefs also show up on the Scheduled Tasks page, so you can see everything your sidekick does on its own in one place.