The agent layer
Agents that work on your data, and only yours.
Give an agent a job, a schedule, a budget and a scope of data it may read. A person approves anything that ships.
One agent
Pricing watcher
Jobwatch a competitor’s pricing page
Scheduleevery night
Budgeta monthly cap, in dollars
Data scopeone brand, the Marketing context
An agent carries your company’s badge and no other.
An agent cannot see anything outside the organization it belongs to. It reads the brands and modules it has been granted, and nothing else. No shared pool, no learning across customers.
This is the difference between an agent and a chatbot with a plugin. The agent is a worker inside your company.
Four controls on every agent. None can be left blank.
Every agent, whether it came from the catalog or you built it yourself, carries the same four controls.
A job
What it does.
Watch a competitor’s pricing page. Flag a deal that has gone quiet. Write the Monday briefing.
A schedule
When it runs.
Overnight, weekly, on a trigger, on demand. A watcher that checks every day costs nothing on the days nothing changed.
A budget
What it may spend.
A monthly cap in dollars, set per agent. When it is spent, the agent stops and says so.
A data scope
What it may read.
One brand or several, one module’s data or the whole context. An agent built for Sales does not read Marketing’s drafts.
Creating an agent is a right, not a default.
Some roles can define agents
and set their schedule, their scope and their budget.
Some roles can run agents
that others built, on the schedule and scope they were given.
Some roles can only read
what agents produce.
For a growing company
The founder can build an agent on a Sunday night, and the new hire cannot accidentally build one that reads the whole pipeline.
For a large organization
IT can decide exactly which three people get to create agents, and the ledger will show who created what.
Roles are documented on Architecture and Security and data.
Agents do not run loose.
They run inside a harness that the product enforces and you configure.
Four scopes of settings
Global, organization, brand, user. Settings inherit from above and can be overridden field by field.
A per-agent monthly budget
Separate from and beneath the platform caps. When it is spent, the agent stops.
The ledger
Every run recorded with who ran it, what it did, when, which model and the exact cost.
The human gate
Nothing an agent produces is published, launched or spent without a person’s approval.
Grounding
Answers quote real figures and name sources, or say the data does not exist.
The agent proposes. Someone signs.
Three kinds of agent are live across the modules.
Watchers
They run on a schedule and look for change.
Competitor pages, search positions, deals gone quiet. When something moves, it lands in the findings feed. When nothing moves, they cost nothing.
Drafters
They produce work on demand or on a schedule.
Articles from market-gap evidence, follow-ups shaped to the deal, briefs scored against your approved content.
The analyst on call
Ask Intelligence answers from your live numbers.
And the weekly operating review tells you what changed and what to do next week.
Give it a schedule and a budget. Read what it finds on Monday.
Roadmap, not built.
Each of these is in the catalog marked not built, and none is sold until it is.
- Rank-change diagnosisNot built
- Content decay predictionNot built
- Budget reallocation with projected impactNot built
- GEO gap asset generationNot built
Six things you can check.
Build the first agent for the job nobody here has time for.
Give it a schedule and a budget. Read what it finds on Monday.
No subscription. No seat fee. Money in a wallet, spent on tokens, storage and data, and your balance never expires.