A custom agent in Agent Hub is how you turn a recurring piece of analysis into something that runs on its own. This article walks through building one end to end: configuring it, putting it on a schedule, running it on demand, and checking what it did.

For what an agent is made of and how its access model works, see What Is Agent Hub?

You need to be an admin on an account with Tooli enabled.

Under Access, choose the roles the agent holds. The agent acts as its own user, and these roles bound what it can see and change on every scheduled and run-once run.

Give the agent the narrowest role that still lets it do its job. A reporting agent needs Reader and nothing more. This is a far stronger guarantee than writing "do not change anything" in the instructions, because it is enforced rather than requested.

Leave Datascope on its default to inherit the datascope from the roles you picked.

The Instructions box is the agent's system prompt, loaded on every run. A prompt that works well usually covers four things:

Be specific about the output. An unattended run has nobody to ask a follow-up question, so anything you leave open the agent will decide for itself.

Optionally add starter prompts, which appear as one-tap suggestions in chat. They do not affect scheduled runs.

Under Tools, set what the agent may do. Use the group menus to set Read-only tools and Write / delete tools in one move, then adjust individual tools.

For a reporting agent, set the write and delete group to Blocked. Remember that Needs approval does not stop a tool during a scheduled run, because there is nobody there to approve it.

Under Skills, leave the access mode on Auto unless your organization has more skills than one agent can carry, or you want tight control over which procedure the agent follows. In either case switch to Pinned and select deliberately. Check the Resolved at runtime list to confirm what the agent will actually receive.

Click CREATE AGENT when the configuration looks right.

Open the agent and switch to the Triggers tab. Each trigger has its own toggle, so chat can stay live while a schedule is paused.

Chat controls whether people can talk to this agent interactively. It runs as the invoking user, under their access.

To add a schedule, click Add schedule.

Schedules run on Toolio's 15-minute tick, in your organization's time zone, so a run starts at or shortly after the time you set rather than to the exact second.

You can add more than one schedule to the same agent. A daily check and a fuller weekly readout can share one set of instructions and differ only in their message.

You do not have to wait for a schedule. Open a custom agent and use RUN NOW, which offers two options:

Either way you can supply an optional instruction for that one run, or leave it empty to use the agent's own instructions. Run-once runs use the agent's roles, the same as scheduled runs.

The Activity tab holds Run History: every autonomous run from the last 30 days, with older runs ageing out automatically.

Each row shows the run's status (Completed, Failed, or Running), what started it (Scheduled, Run now, or Queued), when it started, how long it took, and a one-line summary. For a scheduled run, the detail line names the cadence that fired it; for a manual run, it names the person who started it.

Expand a row and click Open transcript to read the full conversation, including the tools the agent called. When a run fails, the row carries the failure reason.

Run History is the place to check first when an agent is not producing what you expect. The transcript usually shows the answer, whether the agent chose a different tool than you assumed or read a wider slice of the plan than you intended.

Check the status chip first. Disabled means the agent is off entirely, and Paused means every schedule is switched off. Then confirm the individual schedule's own toggle reads Live, and that the time you set has actually passed in your organization's time zone.

Most likely one of three things: its roles do not permit the change, the tool it needed is set to Blocked, or the schedule's Allow write / delete tools toggle is off. All three are enforced independently of the instructions.

Not directly. Create a new agent and paste the instructions across. It is worth reviewing the tool policies and roles as you go rather than assuming the original had them right.

30 days. Older runs age out on their own, so export anything you need to keep.

No. Tooli runs in chat only. Scheduling, RUN NOW, and run history are available on custom agents.