intelligencesupport

Organization and personal agents

Admins tune shared agents and everyone manages personal ones: switching on and off, instructions, model, schedule, budgets, recipients, and test runs on real brand data.

For
Admins and brand managers
Updated

Once an agent is added, you set it up from two tabs of the Agents area. Admins and brand managers work on the shared agents in the Organization tab. Everyone manages their own agents in the Personal tab. Both use the same cards and the same form.

Personal and Organization tabs

Personal holds your own agents. It opens on your personal scope and shows no scope picker. The list starts empty until you add something.

Organization holds the shared agents. Only admins and brand managers see this tab. At the top, the Configure menu picks what you are editing:

  • Organization: followed by your organization’s name and “(shared with everyone in it)”. The agents every member uses. Admins only.
  • Brand: followed by a brand name. The tuning of the organization’s agents for that brand. Admins and brand managers, for any brand of the organization.
  • User: followed by a member’s name and email address. That member’s personal agents, managed on their behalf. Admins only.
The Organization tab, with the Configure menu on top and one card per shared agent
The Organization tab, with the Configure menu on top and one card per shared agent

Who may configure which scope

ScopeAdminBrand managerEditorViewer
OrganizationYesNoNoNo
One brandAny brand of the organizationAny brand of the organizationNoNo
Your own personal agentsYesYesYesRead-only
Another member’s personal agentsYes, in your organizationNoNoNo

Viewers see how agents are set up but can’t change anything. A note at the top of the page explains why the controls are locked. Brands are explained in brands. See roles and rights.

The agent card

Each added agent gets a card. Next to its name and catalog code you may see a Custom badge, or Standard · added by the platform for a standard agent the BearingBridge team added. Then come:

  • A status chip: on · every 20h (with its interval) or off. On the organization scope, an agent the organization left off but some brands switched on reads on for 2 brands (with the real count).
  • edited at this scope when this scope changed any of its settings.
  • What it does and which model it runs on (Calls no AI model for an agent that runs none), followed by what it reads.
  • On the organization scope, the brands it actually runs on, with (brand setting) next to brands that switched it on themselves.

If an agent is on but no brand has its module switched on, the card says On, but no brand qualifies: each brand needs the underlying module switched on.

The On / Off menu on the right switches the agent at this scope. On the brand scope it also offers Inherit, which follows the organization.

Adding an agent here

Under the list, unfold Add an agent to see the standard agents this scope doesn’t have yet. Click Add to open the agent form, filled with the standard version, then Add the agent. It joins the list switched on.

On the Personal tab, Your organization’s agents lists the agents your administrators built for the whole organization; a standard agent you already have isn’t listed twice. View details shows one in full: what it reads, its model, how often it runs, whether it searches the web, and its instructions. Add to my agents opens the same form, filled with the organization’s own name, instructions, model, schedule, budgets and recipients, so you start from what your team already tuned. Confirm with Add to my agents. You then change it, run it and plan it as yours, its runs are billed to you, and the organization’s agent stays as it is. Viewers can open View details but can’t add.

Your own agents can also be copied. On the Personal tab, a watcher or a custom agent of yours carries Duplicate: it opens the form filled with that agent’s setup, under its name followed by “(copy)”. Change what you need and click Create the copy. The copy joins your custom agents, reading the same data; the original stays as it is. Report agents can’t be duplicated.

Anyone who can write can also add a standard agent to their own agents from the catalog, and build one from Create an agent from scratch at the top of the page. See custom agents.

Editing an agent

Click Edit & test on a card. The form opens under it: the agent’s Name on top, then four tabs. The form reopens on the last tab you used on the page, so a save never sends you back to the first one.

An agent open for editing, with its name above the four tabs and the General tab showing what users are told it does and its instructions, each with Improve with AI
An agent open for editing, with its name above the four tabs and the General tab showing what users are told it does and its instructions, each with Improve with AI

General

FieldWhat it sets
NameWhat the agent is called everywhere, including in run emails. Above the tabs.
What users are told it doesThe description shown on the card.
InstructionsAdded to the agent’s prompt. At least 20 characters.

Your instructions add to the agent’s standing rules; they don’t replace them. What it may report and how it grades severity stay fixed.

Each of the two boxes has an Improve with AI button. With text in the box, it rewrites it into a clearer sentence or tighter instructions; with the box empty, it writes it from the agent’s name and scope. It writes with the model picked at the top of the tab, under Improve with AI writes with. The price of each rewrite shows under the box, with Undo, which puts back what you had. Nothing is saved until you save the agent.

Scope

The Scope tab shows everything the agent works on, and every part of it can be changed at any time. Each agent starts from the scope it was built with.

The Scope tab of an agent: its brands, its connections, the modules and features it reads or changes, and the AI it runs
The Scope tab of an agent: its brands, its connections, the modules and features it reads or changes, and the AI it runs
RowWhat it sets
BrandsOrganization and personal scopes only. Each brand, in a run of its own, or One run over the brands ticked below. See none, one or several brands.
ConnectionsThe connections it works on, each marked as yours or the organization’s, with a list: Off, Reads or Reads and changes. For your mailboxes, tick the ones it may use; with all ticked, it uses every mailbox you connect, later ones included. With none to give yet, a link sends you to My Connections.
Modules and featuresOne line per module or feature it works in, named as in the rights grid, each with Off, Reads or Reads and changes. Add a module or feature gives it one more.
Data it readsThe data blocks it is given. A standard agent’s data is shown for reference; a custom agent’s can be changed. See custom agents.
AIThe kind of model it runs: No AI model, A text model, An image model or A video model. Then the model itself, and the Web search switch. With No AI model, a run costs nothing.

A run works inside the scope as saved: it never reads or writes anything the scope leaves out. Switch every module or connection an agent needs to Off and its next run ends on “Nothing in its scope”; take the model away from an agent that reasons with one and it ends on “No AI model in its scope”. Both runs cost nothing, and each tells you what to switch back on.

You can give an agent only what you can use yourself: the modules and features your own rights open, your own connections, and the systems your organization connected. The administration pages are never part of an agent’s scope.

Schedule and budget

FieldWhat it sets
Runs every (hours)The minimum time between two scheduled runs for a brand, from 1 to 720 hours.
Monthly budget (USD)The most it may spend in a calendar month. Empty means no limit of its own.
Weekly budget (USD)The most it may spend over any seven days.

Under the fields, This month and Last seven days show what the agent spent.

Notifications

FieldWhat it sets
Email the results toThe organization’s administrators (on a personal agent, The agent’s owner), These people only (then list the addresses), or Nobody. Only scheduled runs send an email; when you run the agent yourself, the results show on screen.
Email contentSummary and link, or Full results, which puts the findings or the report in the email. Both include a link to the run.

Click Save when you’re done. The button turns amber while you have unsaved edits, on any tab, and changes apply from the next run.

On the organization and personal scopes, an empty field means the standard version of the agent. On a brand scope, an empty field inherits from the organization, and the menus offer Inherit.

Removing an agent

The second button under the form takes the agent back out, and its label tells you what it does. On the organization scope it reads Remove from the organization; on a personal scope, Remove from these agents. After you confirm, the agent’s settings at this scope are cleared and it stops running there. You can add it again at any time, from the Add an agent list or, for your own agents, from the catalog.

On a brand scope the button is Reset to inherited, which sends the brand back to the organization’s settings. For a custom agent it is Delete this agent, covered in custom agents.

Models

That covers the form itself. The next three sections go deeper on three of its settings.

The model is picked under AI, on the Scope tab, with the same model picker as the rest of the app. It lists the models your organization allows for the kind of model the agent runs, each with what one run costs: a typical run for a text model, one image for an image model, one 8 second clip for a video model. Standard keeps the agent’s default (Inherited on a brand scope, which follows the organization). Watchers and custom agents run on Claude Haiku 4.5 by default, a fast model suited to frequent runs. The two report agents, the weekly operating review and the executive report, use the model your organization set for Ask Intelligence until you pick one here.

A practical rule: watchers run often and read a lot, so a small, fast model keeps their cost down, while the report agents write longer documents that can be worth a stronger model. Test both ways on a real brand and compare the cost printed with each result.

Whatever you pick here wins over the default. If that model later leaves the app, the agent goes back to its default. If your organization forbids it, the agent runs on the organization’s Ask Intelligence model instead. See AI models and AI models in administration.

Schedules

Each standard agent comes with a default interval: 20 hours for the search rankings and Google Ads watchers, 44 hours for the AI visibility watcher and the market monitor, 168 hours for the weekly operating review, and 720 hours for the executive report. Custom agents start at 24 hours.

Think of the interval as a minimum gap. An agent is due on a brand again once that much time has passed since its last completed run there. A brand scope can set a different interval for its own brand.

Scheduled runs come from the organization’s agents and from each member’s personal agents, which run on their own schedule as their owner and are billed to them. How the daily schedule picks up due runs, and why a run can fail, is covered in runs and schedules.

Budgets

Each agent carries two budgets, monthly and weekly. Whichever is reached first stops the agent until the window clears: its scheduled runs fail with a message that gives the amount spent, and your own runs and tests are refused with the same message. The form shows what the agent spent this month and in the last seven days.

An organization or brand agent spends the organization’s money. A personal agent spends its owner’s.

When several scopes set a budget, the tightest one wins. A brand or a person can narrow the organization’s budget, never widen it. Runs and tests you start yourself must also fit the spending caps of your account and your organization. See credits and costs.

Testing on real data

Every open card has a Test run box.

  1. Pick a brand.
  2. Click Run on real data.
The Test run box on an open agent card, where you pick a brand and run the agent on its real data
The Test run box on an open agent card, where you pick a brand and run the agent on its real data

The agent runs on that brand’s current data with the name, instructions and model now in the form, even if you haven’t saved them. The result shows below: the number of findings, the model, the tokens, the billed cost in your organization’s currencies, and each finding. There’s no estimate before you click; the cost comes with the result. When the brand has no data for the agent yet, the result says so, and still shows a cost if reading a connected system was charged. For a report agent, the report itself is shown. Open The data the agent was given to see exactly what it read.

Nothing from a test lands in the brand’s feed or among the stored reports. The agent’s comparison snapshot isn’t touched either, so the next scheduled run compares against the same data as before.

Important: A test is still a real model call. It is billed like a run and checked against the agent’s budgets.

The brand must belong to the organization the agent belongs to, and the agent’s module must be on for it.

An agent set to read several brands is tested the way it runs: once over all the brands ticked in the form, saved or not, whatever brand the test box shows. With no brand ticked, the test reads the organization’s own data.

Sometimes the answer is that nothing is worth reporting. That’s a fair result. If you expected findings, tell the agent more precisely what counts as one, then test again.

Reading reports

Once opened with Edit & test, the card of a report agent has a Recent reports box. Pick a brand to read the latest reports stored for it. The first report arrives on the next scheduled run.

Troubleshooting

If a save or a test is refused, the message under the card tells you why.

MessageWhat to do
The instructions need at least 20 characters. Clear the field to inherit instead.Write a longer brief, or empty the field to use the standard one.
The interval must be a whole number of hours between 1 and 720.Enter a whole number in Runs every (hours).
List at least one address, or send to the organization’s administrators.With These people only, add at least one email address.
Not an email addressFix the address named in the message.
Agent needs its module switched on for brand.Pick another brand for the test, or ask an admin to switch the module on for that brand.
Pick a brand of the organization this agent belongs to.Test on one of your organization’s brands.
No organization, brand or member scope to configure. Your own agents are under Personal.Your role gives you no shared scope. Use the Personal tab.