Chatting and running missions

Two screens cover the day-to-day work: Chat for a conversation, Missions for larger multi-step jobs.

Chat

The Chat screen is a direct conversation with your assistant — type a message, get a reply. As it works it may use tools (reading a file, searching the web, writing to memory); you’ll see each tool call as it happens.

A few things worth knowing:

It streams. Replies appear as they’re generated.
You can cancel. If a turn is taking too long or heading the wrong way, cancel it and try again.
It remembers. Within a conversation it keeps context; across conversations it draws on its long-term memory (see the Memory page).
Risky steps pause. If a reply requires something irreversible — sending a message, deleting a file — the assistant stops and asks you to approve first.

When it asks first

Some actions can’t be undone — deleting or overwriting one of your files, committing to a git repository, sending an email, placing an order. When the assistant wants to do one of these, it pauses and asks you:

In the Studio, an Approval needed card appears in Chat with what it wants to do, why it’s asking, and the exact details (under Arguments). Click Approve or Deny.
In the terminal, it shows the same thing and asks Approve? [y/N]. Type y to approve; Enter (or anything else) says no.

If you approve, it does exactly what it showed you and carries on. If you deny — or don’t answer within 10 minutes — it doesn’t, and it tells you so. The assistant can never approve on your behalf.

In the chat apps (Telegram, Discord, Slack) there’s no card: the assistant tells you what it wants to do and stops, and replying “yes” in your next message lets it go ahead once.

If you press Ctrl-C while the terminal is asking, the reply is cancelled; press Enter to get back to the chat.

Missions (orchestration)

A mission is a larger goal you hand off — something that takes several steps, or several specialists working together. The Missions screen shows each mission’s plan, progress, and any points where it’s waiting on you.

How a mission works:

You state a goal.
The assistant breaks it into a plan of steps.
It works through the steps, in order or in parallel where it can.
When a step needs your sign-off (an approval gate), the mission pauses and surfaces the decision. You approve or reject, and it continues.
When it’s done, you get a summary of what happened.

Missions are where Aivyx PA’s teams come in — for a complex job, a lead assistant can delegate steps to a crew of specialists, each with only the access its job needs. See the Teams page for how that’s set up.

Approval gates

Gates are the heart of the safety model for autonomous work. The assistant can do all the analysis and preparation unattended, but it stops at the consequential step and waits for a human. You’ll see gates both here and in Chat whenever an action crosses that line.