Symbol Docs
Features

Inbox

Your notifications, from security alerts to updates from creators you follow, and the questions an AI assistant is waiting on you to answer.

Every account has an Inbox, and it is part of Symbol for everyone. It holds two kinds of thing:

  • Notifications: something changed in your account's security, a creator you follow published something new, or someone shared something with you.
  • Agent messages: an AI assistant needs a Decision only you can give. Agent messages are experimental and need the Experimental features toggle.

Notifications for everyone

Your notifications are listed newest first. The Inbox link in the sidebar shows how many you have not read yet.

  • Security notifications. Each security email Symbol sends you also appears here, saying what happened and when, with a link to Security settings. Reading or clearing one never stops its email.
  • New work from creators you follow. When a creator you follow makes something public, a notification links to it. It arrives within about fifteen minutes of the creator publishing, and a creator's item notifies you once: if they take it private and publish it again, you do not get a second one.
  • Things shared with you. When someone shares a capsule, Type, Collection or project with you by name, a notification links to it.

How notifications behave:

  • Opening a notification marks it read. Mark as read clears one without opening it.
  • A notification is checked again every time you open the Inbox. If you can no longer see the item, because it was made private, removed, or your access ended, it says "This item is no longer available" and shows nothing else about it.
  • A notification never grants anything. Opening one takes you to the item or the setting; what you can do there depends on your access at that moment.
  • Security notifications show only in your own browser session and in the Symbol app. An API key, a connected app, an organization key or an AI assistant connected over MCP never sees them, and they are not counted in its unread total.

You do not need the Experimental features toggle to see notifications. Turning it on or off changes only agent messages.

Agent messages and Decisions

Agent messages are experimental. Turn on Experimental features in Settings to see them. Turning it off hides them again without deleting anything, and they come back when you turn it on; your notifications stay either way.

Most of the time an assistant works without you. Sometimes it needs a decision only you can make: which option to take, whether to change something live, or a fact it cannot find in your knowledge base.

Those questions land in your Inbox too. An assistant sends one, keeps working, and picks up your answer when you give it.

How a Decision looks

A Decision is a short structured form, not a wall of text. Each question is capped at twenty words, and a choice question is answered with a tap.

They arrive one question per screen. Answer the required ones and submit. Some Decisions carry an optional bonus section behind an "Answer more" gate, which never blocks submitting.

Answers are write-once. Once you submit, the Decision becomes read-only. You can see what you chose, but you cannot change it. If the assistant needs a different answer, it has to ask again.

When an assistant stops waiting

An assistant that asked you something may not wait forever.

  • If it is still checking for your reply, the Decision shows as waiting.
  • If it gave up and carried on with its own judgement, it withdraws the Decision. A withdrawn Decision goes inert: it stops showing as waiting, and a later reply is refused.

The Inbox distinguishes these two states, so a question that no longer matters does not sit in your list looking urgent.

Snoozing

Snooze a waiting Decision to take it out of your unread list. It wakes later and returns to unread. Use it when the question is real but not now.

Replying without a form

You can also send a free-form message back into a conversation. When nothing is waiting on you, the box tells you so instead of implying there is a question to answer.

If a send fails, your unsent text stays in the box. It is not discarded.

Conversations and identity

Messages from the same assistant thread into one conversation, including across separate sessions. An assistant that has been packaged with a picture shows that picture; one that has not gets a default.

Mute a conversation to silence its unread badge. Archive it to hide it from the list, and use the archived filter to bring it back.

Browser notifications for Decisions

The first time a Decision arrives, your browser asks for notification permission.

  • Allowed: new Decisions raise a notification, and one arriving in a tab you are looking at plays a sound.
  • Denied: no notification is raised. The unread badge still updates, so nothing is lost, you just have to look.

Context on a question

An assistant can attach its own reasoning to a question: what changed, which value it is asking about, the reference it is working from. That appears in a Context panel beside the question.

When an assistant supplies its own context, it replaces the generic summary rather than sitting alongside it.

Agent messages and Decisions are experimental and require the Experimental features toggle in Settings. Notifications do not.

On this page