Your agent
Power, autonomy, how much it may run per day, and how to ask for work in a way that produces something usable.
The agent is the part of BizOS that does the work. This page is about steering it.
Power — on or off
Power is the master switch. Off means no autonomous run happens at all: the agent will still answer you in chat, but it will not go off and work on its own.
Turn it off when you want the company to stand still — during a holiday, or while you rethink your positioning.
Tasks per day
Tasks/day caps how many autonomous tasks the agent may run in a day. It is the dial most people actually use day to day:
- Low — the agent picks only what matters most. Good while you are still learning to trust it.
- Higher — more ground covered per day, more Work Credits spent per day.
Setting it to off is equivalent to zero autonomous tasks: the agent then only works when you explicitly ask.
Autonomy on spending
Separately from how much it works, you decide how far it may go on money:
| Mode | What happens |
|---|---|
| Ask before spending (recommended) | Everything runs on its own — only ad spend waits for your OK. |
| Full auto | Ads launch and adjust alone, within your daily cap and wallet. |
Even on full auto the agent cannot exceed the daily budget and wallet you configured. Autonomy changes who clicks "go", never the ceiling.
Talking to it well
The agent is good at executing a clear intent and poor at guessing an unclear one. Three habits that change the output:
Name the deliverable. "Write X and save it as a document" beats "think about X". A deliverable can be judged; an opinion cannot.
Give the constraint. Audience, tone, length, what to avoid. The constraints you leave out are the ones it will get wrong.
React to real output. Correcting an actual document is worth more than a long briefing. It learns your standards from your corrections.
Slash commands
| Command | What it does |
|---|---|
/goal | Turn an outcome into action |
/next | Set the next tasks |
/ask | Ask without taking action |
/brief | Daily progress and blockers |
/report | Create a saved report |
/review | Review an asset or result |
/support | Contact support |
/ask is the one to remember: it answers without starting work, so it never
spends a Work Credit.
Attachments
You can attach images or PDFs to a message — a brand guide, a competitor's page, a spreadsheet to work from. Treat anything you attach as material for the agent, not as instructions it must obey blindly: it reads them as data.
What it will refuse
The agent will not act on a capability that is not connected. If LinkedIn is not linked, it will not promise a LinkedIn post; if outbound email is off, it will not claim to have sent anything. When something looks impossible, check Connections before assuming a bug — and if you disagree with a refusal, the FAQ covers the common causes.