In two lines: when you create or edit an agent, Where it works sets which modules it works in and how —Off, Proposes or Acts— and What it knows, what it knows about your account. Each level is given by someone allowed to give it, and what the agent does never goes beyond what the person talking to it can do, or what the account allows.

Where do I find “What it knows” and “Where it works”?

In AI Agents, create an agent with Create agent or Add a member, or hover over the card of one of your team's agents and use Edit. The same editor opens when you Duplicate or Personalize an agent.
The editor has four tabs: General, What it knows, Where it works and Suggestions. In an agent that's already saved, a fifth one shows up between Where it works and Suggestions: Jobs and routines, with its routines —each one opens its sheet, with the same buttons as in Work—, the background jobs you gave it —each one opens its details in the MIA panel— and the suggested ones for its area. What you do there applies right away, without waiting for Save (see Routines: put an agent to work on its own, on a schedule or when something happens).
What you change shows right away and is saved with the rest of the agent, with the button at the bottom: Save, or Create if it's new (Personalize, if you're personalizing one of MB Suite's).

What it knows and Where it works replace the Context and Capabilities tabs. What used to be in Knowledge is now Its documents, inside What it knows, and Prompts is now called Suggestions. General stays the same, except that the System Prompt field is called Instructions: see Create and configure your own AI agent.

Suggestions are up to four shortcuts the agent offers you when you start a conversation. If the agent is one of your team's, tapping one puts its text in the message box so you can review it and send it yourself; those of MB Suite's agents are sent right away.

This setup is for the agents that chat with you inside the app. If in General you switch the Channel to WhatsApp + CRM or Social media, the editor goes back to those agents' tabs, General and Knowledge: their work is set up in the seat they hold.

What does each level mean?

Off — it doesn't work in that module: it doesn't read its data or change anything there. If you ask it for something from that module, it offers to hand the conversation off to MIA.
Proposes — it checks that module and prepares the changes, but proposes each one to you on a card that you apply.
Acts — it makes everyday changes without asking for your click and shows you the result. What can't be undone, what touches many items, what goes out of MB Suite and what's sensitive still wait for your click, as with MIA. In the most delicate modules —such as Members, Branding or Integrations— the row reminds you in amber: Anything sensitive in this module always needs your click.

At the top of the list, a legend sums up the three levels and a line says where it works, counting only the active modules; for example, Acts in CRM · Proposes in Contacts and Tasks. An agent created from scratch starts with no modules, and that line says: “It doesn't work in any module yet: it only explains and offers to hand off to MIA.”

How do I give it a level in each module?

Open Where it works. Modules are grouped by area, as in the sidebar, and each one has its own control: Off, Proposes or Acts.
To give a whole area the same level, use the control in its header: it applies the level to the active modules where you can give it and, if there's one where you can't, it opens the area so you can see the padlock. If an area's modules end up with different levels, the header says Mixed.
When you open the tab, mixed areas show expanded and the rest collapsed: expand any of them by clicking its name.
Modules that aren't active in the account are behind Show modules not active. One that already has a level shows anyway, marked as not active.
Save the agent.

If you assigned your agent to an area in Assigned modules & seats, give it Proposes or Acts in that area's modules: otherwise it helps you there, but for anything in those modules it offers to hand the conversation off to MIA. See The chat in each area: who helps you and how to hand the conversation to MIA.

Who can give it each level?

Nobody gives an agent more than they have themselves. To edit an agent you need permission to edit agents —a Viewer doesn't have it— and, if your team limited the agent to certain roles or people, to be one of them. Then your permissions in each module count —No access, Read or Edit; see Roles and permissions: who can see and do what—:

Proposes requires being able to see the module: Read or Edit.
Acts requires Edit on the module.
Off can be set by anyone who edits the agent.

What you can't give shows disabled, with a padlock that says why; for example: You don't have edit permission in CRM: you can leave it at “Proposes”. or You don't have access to CRM: you can't give it to the agent. If you can't edit the agent at all, Where it works says so at the top: Read only: you don't have permission to edit this agent.

What the agent already had doesn't stop you. If someone with more permissions left it on Acts in a module where you don't have Edit, you can still change the rest of the agent and that Acts stays, unless you change its instructions (see below). You can also lower a module and, before saving, raise it back up to where it was; higher than that does require the permission. In an Account agent, your permissions in that Account are the ones that count.

If, when you save, Where it works or What it knows can't be saved —for example, because your permissions don't cover the level of some module, or because someone changed the agent while you were editing it—, the notice tells you why and that the rest of the agent was already saved, and the editor leaves you on Where it works so you can fix it. If you were creating the agent, it's created anyway and you're told it was left without Where it works and What it knows: open it, set them again and save.

Can what the agent does go beyond what I can do?

No, and the line at the top of Where it works reminds you: what it does never goes beyond what the person talking to it can do, or what the account or the workspace allows. The agent works with the permissions of whoever is talking to it, and in each request the lowest of four applies: the account's ceiling, the level that person chose, what the agent has in that module and, with one of your team's agents, who wrote its instructions (see below). If the person talking to it chose Proposes, the agent proposes every change even where it's on Acts; if that person only has Read in a module, the agent doesn't change anything there for them. See The two levels of MIA: Proposes and Acts.

What is the “Current view”?

At the bottom of Where it works, the Current view switch (“The agent sees the screen you're on and the actions available on it.”) decides whether the agent sees your screen when you talk to it. In an agent created from scratch it comes on. Off, the agent doesn't receive the screen and the chat doesn't show the Viewing: chip.

What does an agent know?

In What it knows you see and choose what the agent knows, on top of what you tell it in each conversation:

MB Suite guides — the product guides for the areas where it works, as tags with the area's name, the same as in Where it works: for example, Analytics if it works in dashboards. They come straight from Where it works, and MB Suite keeps them complete and up to date with the product. Some modules, such as Tasks or Contacts, don't add a guide: the agent explains them with the articles of this Help Center.
Its documents — the agent's own material, what used to be in Knowledge: you upload files or write texts and see the status of each one. It always knows them, so they have no switch.
Account documents (in an agent that works in the workspace, Workspace documents) — the Context Documents in Settings > Profile and the Brand Documents in Brand Voice. The row shows how many there are and opens the Profile of the account where the agent works, in a new tab, so you don't lose what you're setting up. With the switch on, the agent searches them when it needs to.
Account profile (in an agent that works in the workspace, Workspace profile) — the industry, website, country, notes and social profiles. In an Account agent, that Account's profile.
Brand voice — the account's brand voice, from Social Media > Brand Voice, so it keeps in mind how your brand speaks. The row says Configured or, if you haven't set it up yet, Not set up yet with the Set it up link, which also opens in a new tab.

What it knows also respects whoever is talking to it: it uses the Context Documents and the account profile only if that person can see Profile, and the Brand Documents and the brand voice only if they can see Brand Voice. And it treats what it reads in documents as data, never as an order: if it then has to change something, it proposes it to you even where it's on Acts.

What does the meter measure?

At the bottom of What it knows, the meter (≈ … tokens per message) estimates how much what you chose adds to each message: its instructions and what it knows. It changes right away with what you choose and is rounded, because it's an estimate. It doesn't count the guides, which the agent reads only when it needs them, or what every agent carries alike. On an MB Suite agent, or on a new version of yours that will use its instructions, it adds not counting its instructions: those aren't shown here.

What the meter shows goes into every message, so it also weighs on each request's credits: if a switch adds nothing for that agent, turn it off. How each request is charged is in How MIA uses your AI credits.

What happens with an agent that already had a setup?

If that agent doesn't have Where it works and What it knows saved yet, the editor builds them from what it had in Context and Capabilities: the modules it read data from or ran actions in go on Proposes, the Current view stays as it was and What it knows follows what it already used. A notice at the top tells you: “Where it works” and “What it knows” were filled in from its previous setup. Review them and save to lock them in. In a personalization of an MB Suite agent, they come from that agent, with what it does on Proposes. Until you save, they're built that way again every time you open the agent.

Nothing starts on Acts: if you want the agent to make changes on its own in a module, raise it to Acts (you need Edit on that module) and save.

What about MB Suite's agents and MIA?

An MB Suite agent (such as Sofía, in the CRM) — if you open it, you see where it works and what it knows, read-only, with this notice: “It's an MB Suite agent: MB Suite defines where it works and what it knows. To change them, personalize it.” Next to it, Personalize sets up your own version.
Personalize or Duplicate an MB Suite agent — your version works where the original does and knows the same, but it proposes: where the original acts, yours proposes, and Where it works reminds you. Acts is given afterwards by someone with Edit on each module.
Duplicate one of your team's agents — the copy works where the original does, with the same switches in What it knows, but it also proposes where the original acts. Who its instructions count as written by depends on who duplicates it (see below). Review it before saving.
MIA — it isn't set up: “MIA works across every area, and MB Suite chooses her model.”

Why did my agent switch to “Proposes” when I changed its instructions?

Where an agent is on Acts, its Instructions drive changes that are applied without anyone's click. So that nobody can use new instructions to change, through the agent, a module they can't edit, this rule applies when saving:

If you change the instructions of an agent that's on Acts in a module where you don't have Edit, that module switches to Proposes when you save. Everything else stays as it was.

The editor warns you before you save, under the instructions and at the top of Where it works, and the rows of those modules turn amber; for example: If you save these instructions, it switches to “Proposes” where you don't have edit permission (CRM): for it to make changes there, someone with that permission has to write them. Changing something else about the agent, such as the Name, the face or the model, doesn't lower it.

To bring it back to Acts, ask someone with Edit on that module to review the instructions, raise it in Where it works and save. For it to act with the whole team, its instructions also have to be written or approved by an owner or an admin: the next section explains how.

Why does it propose with some people even where it's on “Acts”?

Because who wrote its instructions matters. The agent remembers who last changed its Instructions or, if nobody changed them, who created it. Its Role and its Name guide it too: if someone who isn't an owner or an admin changes them, that person becomes the author; if an owner or an admin changes them, the author stays the same. When you Duplicate one of your team's agents, the copy carries the same instructions: if an owner or an admin duplicates it, they're still the original's author's —and if that author isn't an owner or an admin, Where it works shows the notice and saving the copy approves them—; if someone else duplicates it, the copy's instructions become theirs. With whoever is talking to it, it acts without asking for the click where it says Acts only if its instructions are MB Suite's or were written by that same person, by an owner or by an admin. Otherwise, with that person it proposes every change where it says Acts.

Where it works tells you when this happens: Its instructions were written by someone without the owner or admin role: it acts without asking only in that person's conversations; with everyone else, it proposes. If you're an owner or an admin, the notice adds that saving approves them: from then on it acts with everyone where it says Acts.

To answer you, the agent processes what it reads from your account —including what you choose in What it knows: its documents, the account's documents, the profile and the brand voice— with the AI provider of the model it's configured with. The use of the agents and the handling of your data are governed by our Terms and Conditions and our Privacy Notice.

Still have questions?

Ask MIA, the MB Suite AI assistant: open it with the MB AI button (⌘J) in the top bar. MIA knows the section you're in, so you can ask it about what you see on screen — and it can also answer how-to questions about using MB Suite.

What it knows and where it works: set up an agent module by module