Agent Hub is where your organization's AI agents live. Find it under Settings > AI & Agents > Agent Hub.

An agent is an identity, not just a prompt. It has its own instructions, its own set of tools, its own skills, and its own roles and datascope, exactly like a person on your team. That is what lets an agent run on a schedule while nobody is watching, and still stay inside the boundaries you set for it.

Agents are organization-scoped. Anyone who can reach Agent Hub sees the same list, and there is no personal or private agent.

Agent Hub separates agents into two groups.

Built-in holds Tooli, the general-purpose merchandising assistant that ships with Toolio. Tooli runs in chat as you, with your access. You can change its model, its tools, its skills, and its memory access, but its name, description, and instructions are managed by Toolio and cannot be edited.

Custom agents are the ones you create. A custom agent gets everything the built-in agent has, plus its own instructions, its own schedule, and its own run history. Each row shows the agent's status, the outcome of its last run, its model, and who last updated it.

Open any agent to see its configuration. The Configure tab holds six sections.

The agent's name, description, avatar icon and color, and the model it runs on. Leave Model set to Server default unless you have a reason to pin a specific one; the default tracks whatever Toolio currently recommends.

The description is worth writing carefully. It is what your team sees in the hub list when deciding which agent to run.

This is the section that matters most, and the one worth reading twice.

The agent acts as its own user. It holds one or more roles and a datascope, exactly like a person, and those bound what it can see and change during scheduled and run-once runs. An agent with only the Reader role cannot write to your plans, no matter what its instructions say.

In chat, this does not apply. A chat conversation runs as the person typing, under that person's own access, so two people talking to the same agent can legitimately see different data.

Leave Datascope on its default to inherit the datascope from the agent's roles.

The system prompt. This is the agent's stable identity and behavior, and it is loaded into context on every run. Say what the agent should produce, at what level of detail, and what it should do when the data looks wrong.

Below the prompt, starter prompts are optional one-tap suggestions shown in chat. They apply to the chat trigger only, and have no effect on scheduled runs.

The built-in Tooli agent has no Instructions section, because its prompt is managed by Toolio.

Tools are the atomic actions the agent can take, served by the Toolio MCP server. They are grouped into Read-only tools and Write / delete tools, and each one carries a policy:

Set a policy on a whole group at once with the menu at the right of the group header, then override individual tools underneath.

How a policy behaves depends on how the agent was started, and the difference is important:

Skills are reusable procedures that orchestrate several tools and read reference docs, loaded when they are relevant. Choose how the agent reaches them:

Only organization and system skills can be pinned. Personal skills stay with the member who created them and are never shared onto an agent.

There is a cap on how many skills one agent can carry. If you are in Auto mode and your organization has more skills than the cap, Toolio warns you and serves only the top slice, system skills first. When that happens, switch to Pinned and choose deliberately rather than letting the cut happen for you. The Resolved at runtime list at the bottom always shows exactly which skills the agent will actually receive.

Durable notes about how your organization works, carried into every conversation with the agent. Set Organization memory access to Disabled, Read only, or Read and write depending on whether this agent should be able to add to that shared memory or only draw on it.

Every custom agent shows a status chip in the hub list and in its editor.

Paused is the useful middle ground. Chat keeps working while the automation stops, which is what you usually want during a data migration or a quiet period.

Agent Hub appears only when Tooli is enabled for your organization and you are an admin. If the AI & Agents section is missing from your settings, or Agent Hub is not in it, you are not an admin on the account or Tooli is not switched on.

Inside the hub, the Access card and the trigger toggles are admin-only. Editing the built-in Tooli agent also requires admin.

No. Agents are organization-scoped and shared. Personal skills are the per-member concept; agents are not.

Yes. Deleting an agent permanently removes the agent and its configuration, and the confirmation dialog says so. Export anything you need first.

Because chat runs as the invoking person, under their access and datascope. Scheduled and run-once runs are the ones that use the agent's own roles, so those are consistent no matter who set them up.

Only if you let it. An agent can write only when its roles permit the action, the relevant tool is not Blocked, and, for a schedule, Allow write / delete tools is switched on for that schedule. Give agents the narrowest role that still lets them do their job.