Create an agent from a preset, a plain-English description, or from scratch. The autonomy dial, the confidence threshold, automatic spending limits, the Try it test run that changes nothing, and the anatomy view.
From the Agents home, choose New agent. There are three on-ramps.
The New agent button on the Agents home opens a dedicated creation page. You can copy one of the preset agents, describe what you want in plain English, or start from scratch. Whichever you pick, a short Make it yours check follows: confirm the name, the record type the agent works on, and its brief, with a read-only summary of the schedule and the approval posture. Choosing Create agent then opens the full builder, and the result is a draft: nothing runs until you publish it.
Copy one of the built-in agents and make it your own.
Under “Start from a preset” is the list of built-in agents, each with a one-line description of what it does. Choosing Use as starting point gives you a custom draft with the same instructions and actions as the preset, ready to tailor. You review and edit everything before it runs.
Brief the agent like a teammate and let Laureo draft it.
Under “Or describe it” is a free-text box. Write a sentence or two the way you would brief a new team member, for example “Every morning, review my open deals, flag the ones that have gone quiet, and suggest a next step for each.” Choose Draft it for me and Laureo drafts the agent for you: instructions, a suggested set of actions, and a schedule.
The draft lands in the Make it yours check for you to review and adjust before anything is created. It is never published or run automatically. If your description mentions something the agent cannot map to a real action, the page tells you so you can pick an equivalent in the builder afterwards.
Write the brief yourself and set everything up in the builder.
Under “Or start from scratch”, choose Blank brief. You name the agent, pick the record type it works on, and write a short plain-English brief of what it should do. From there the full build screen opens, where you adjust its schedule, grant actions, and set the approvals and limits described below.
A rail down the side of the builder that shows the whole agent at a glance.
The build screen keeps a persistent anatomy rail that lays the agent out as six stages, in the same order as the sections of the build screen. Each stage summarises how it is configured and jumps to that section when you click it, so you can see what is set and what is still missing. Run conditions and Approvals and limits live inside the Advanced disclosure; clicking their stage opens it for you.
| Field | Description |
|---|---|
| Trigger | What starts the agent, for example a daily schedule or an event. |
| Instructions | The plain-English brief the agent follows, like a brief for a teammate. |
| Tools | How many actions it has been granted, and which ones. |
| Run conditions | Optional filters a record must meet before it is worth a run. |
| Approvals and limits | The autonomy setting that bounds it. Spending is limited automatically by your plan allowance. |
| Output | What it produces: drafts for review, or proposed actions for approval. |
The exact set of things the agent is permitted to do.
In the Allowed actions section you grant the agent its tools, searched and granted by family. The agent can never act outside this list. Each action carries a risk badge, Read, Draft, or Write, and a write action that changes data shows a Needs approval lock. An agent with no actions granted can still read your CRM to reason, but it cannot propose or take anything, so grant at least one.
How much the agent may do before a person is involved.
The Autonomy control sits in the Approvals and limits section, inside the Advanced disclosure. It has three positions, and every agent starts on the safest one.
| Field | Description |
|---|---|
| Suggest only | Proposes every action for your approval. Nothing runs on its own. This is the default and the mode every preset uses. |
| Draft and hold | Prepares drafts and safe reads on its own; anything that changes data still waits for you. |
| Act with review trail | Acts on its own when its confidence clears your threshold, and logs every action. High-risk actions still wait for approval even here. |
How sure the agent must be before it acts without you.
The confidence threshold is a percentage from 0 to 100, set in the Model section inside the Advanced disclosure, alongside model quality. Below the bar you set, the agent routes the action to approval instead of acting on its own, even in an autonomy mode that would otherwise run it. Start high and lower it as you build trust. At 100 the agent always routes its actions to approval.
Model quality sits next to it: pick Light Tasks (the default, quick and inexpensive), Standard for a balanced default, Advanced for nuanced work, or Advanced reasoning for the hardest judgment calls. Higher tiers cost more per run.
Automatic limits that stop a runaway agent, with nothing to set up.
Everything an agent spends counts against your plan’s AI allowance, the same allowance the rest of Laureo’s AI shares. On top of that, every agent carries an automatic limit: a single agent can use at most 25% of your monthly allowance in a month, and at most 5% of it in any one day. A runaway agent is caught within a day rather than at the end of the month.
See exactly what the agent would do on a real record, with zero side effects.
Before you publish, use Try itto run the agent for real against one of your own records. The model genuinely reasons and calls its tools, but every action runs in preview: it builds what it would do and returns without writing. The panel says so plainly: “Test run on a real record. Nothing changes.”
Nothing runs until you publish, and every publish is a restore point.
Choose Publish & activate to make a draft live, or Publish changes on an agent that is already running. Instructions save as you type, but a change only takes effect when you publish. Republishing a live agent asks you to confirm a new version: records already mid-run finish on their current version, and new runs use the new one.
Every publish captures the previous configuration as a version you can restore, so you can always roll back to how an agent was set up before a change.