Tool groups

At the bottom of every agent editor there’s a collapsed Capabilities section with a short list of switches. Each switch is a tool group — a bundle of related abilities that turn on together, rather than a hundred individual tools you’d have to reason about one at a time.

Groups vs. integrations#

These are two different dials and people mix them up constantly:

• Integrations connect an agent to a source: your Google Ads account, an MCP server, a knowledge base, a custom API. Without one, the agent has no data.
• Tool groups grant a class of ability: making charts, writing files, proposing campaign changes. Without the right group, the agent has the data but can’t do the thing.

An agent that can’t produce a chart usually has the data and is missing the Visualization group. An agent that says it has no access to your spend is usually missing the ad-platform integration.

The seven groups#

Memory#

Search your organization’s knowledge and keep private learnings across conversations.

Turn it on when the agent should get better over time, or should be able to look things up that live outside the current conversation. Leave it off for a strictly stateless job.

Folder access#

Browse and edit the shared organization workspace — files, folders, and saved creatives.

Turn it on for anything that reads your briefs, playbooks, or brand documents, or that should file its own output somewhere your team will find it.

Visualization#

Render charts, tables, and KPI cards inside answers.

Artifacts#

Produce PDF documents — in the side-by-side canvas in chat, and as a real file the agent hands downstream when it runs in a workflow node or on a schedule.

This is the group that makes “email me a PDF every Monday” possible: on an automated run the document becomes an attachment the next step can send or post. Not available to sub-agents, Slack, or embedded chat.

Creative Studio#

Generate, edit, save, and launch ad creative variants. Pair it with Creatives work — remixing a winner, filling an angle gap, producing variations off a competitor ad from your swipe file.

Suggestions#

Propose campaign changes as cards your team approves in Home or Slack.

This is the most important group to understand, because it’s the one that makes unattended automation useful and safe:

• The agent never applies anything itself. It writes a pending proposal with its reasoning and estimated impact; a human approves it.
• It is the only route by which a scheduled or background run can lead to a platform change. A direct change needs an approval preview, and a background run has nobody to answer one.

It needs connected ad accounts, plus an ad-platform integration on the agent so it can read performance in the first place.

Data tables#

Read and edit your own data tables — query rows, create tables and columns, add, update, and delete rows.

Your synced ad-performance tables stay off-limits: this group covers the tables you create, not Adlyse’s own record of your platform data.

Where each group works#

Two of the groups depend on something being able to receive what they produce, so they’re only active on the surfaces that can deliver it.

For the two limited rows, switching the toggle on means “wherever this can work” rather than “everywhere” — and switching it off still means off everywhere.

How to choose#

Start with everything off, then add one group at a time in response to something a real run couldn’t do. The order that usually works:

1. Ad-platform integration — almost every agent needs one.
2. Suggestions — if the agent should be able to recommend changes.
3. Folder access — if it needs your briefs, playbooks, or brand documents.
4. Artifacts — if a run should end in a document somebody receives.
5. Everything else, only on evidence.

An agent with three capabilities and one clear job is more predictable, easier to trust unattended, and cheaper per run than one with all seven.