From the Agents home, choose New agent. It opens the agent’s Build page straight away. The name at the top reads Untitled agent until you press it and type one, the Instructions box sits under it, and the three sections that follow are shown as headings with one line under them: Available once you start writing.

Opening the page creates nothing. The agent is saved as a draft the first time you do something on purpose: you type a sentence of ten characters or more, press an example, rename it, press Set up from instructions, or choose Or fill in the form yourself, which opens every section so you can fill them in by hand. Whatever you typed carries over, and from then on the page is the agent’s own Build page.

Under the heading, the page tells you what to cover: Tell it, in plain words:

• What to do, and for which records
• When to run: a time, or when something happens
• Anything it must never do without asking you

The box takes up to 5,000 characters, and a counter under it shows how many you have used. While it is empty, a row of examples sits under it headed Start from an example. Pressing one puts that brief in the box so you can change it, and it sets nothing else: no schedule, no actions and no level. Every example asks only for what an agent can already do on the Business plan, so the agent it describes can be published and run on any plan that builds agents. On the Ultra plan, if you can open Service settings, one line under the examples is a link rather than an example: that work already runs in another part of the product, so the link takes you there and puts nothing in the box. The row goes away as soon as the box has words in it.

Once the box holds a sentence, press Set up from instructions. Setup reads what you wrote and fills in the sections below it: when the agent runs and on which records, and what it can do, plus a name while the agent is still untitled. Each action it chooses that changes something is set to Ask first, and the level to Ask before acting, whatever your instructions say. That holds every time it runs, so an action it chooses that you had allowed, or a level you had raised, comes back down, unless you made that choice by hand since the last setup on this visit (before the first one, since you opened the page). You choose Allow or a higher level yourself. Until you have written enough, the button stays off and the page says Add a sentence about what it should do.

While it works, the button reads Setting up… and one line says what is happening: Reading your instructions, then choosing when it runs and what it may do. Only the box is locked; every other section stays editable. After a few seconds the line adds a running count and a Stop link. If setup cannot finish, one line says why and the form stays exactly as it was. Once setup has run, the button reads Update setup for the rest of your visit, so you can change your instructions and run it again; after Undo setup it goes back to Set up from instructions. Each setup uses a small part of your plan’s monthly AI credits.

When setup comes back, a card titled How I set it up opens under the box. It would act on now is always there; every other part shows only when it has something to say.

Each section setup changed carries the muted mark Set from your instructions beside its heading until you edit it. On the level, and on What it can do when setup changed only the decisions there, the mark reads Set by setup instead, because setup sets those the same way whatever your instructions say. Undo setup at the foot of the card puts every field back the way it was before that setup, and the card then reads Setup undone. The foot of the card also says, once: Uses your plan's monthly AI credits. The card and the marks belong to this visit to the page: reload it and they are gone, while the settings stay.

Everything about when the agent runs lives in this one section. The Starts picker has two groups. On a schedule offers Every hour, Every day, Every weekday, Every week and Every month; pick one and set the time zone, the time (except for every hour), the days for a week and the day of the month for a month. When something happens lists the events Laureo actually sends for each record type, written as sentences about the record, and a field change then asks Which field. An event agent runs once for every matching event, so ten qualifying records in an afternoon is ten runs.

While the agent has never been published, the event you pick sets the record type it works on, and for a schedule Works on sets it. Once it is published, the list keeps to its own record type and the section says Record type is fixed once published.

Under the picker, one sentence reads the whole answer back, in words and never as a schedule expression, for example Runs every weekday at 9:00 AM. Next run: Monday at 9:00 AM. A schedule kept in another time zone names its place, and a paused agent reads Paused. It will not run until you turn it back on. instead of a next run.

For an event, Only for records where narrows which records are worth a run. Fields are grouped so the list stays readable: Basics, Owner, Dates and Amounts. Identity fields and the ones that have no list to pick from are not offered; a condition saved earlier on one of them still applies and shows under Advanced, where you can change or remove it. A record is picked from a list, never typed as an identifier, and a yes-or-no field reads Yes or No. A row with no value yet says Pick a value. and is not saved. Each condition is named in the sentence that reads the section back.

An agent, like a rule, can be built on Companies, People, Deals, Projects, Tasks, Activities, Tickets, Quotes, Products, Campaigns and Goals. An agent can also start on inbound email: under When something happens, choose An email arrives. It then runs once for each email, and it reads only the mailbox of the teammate it runs as, never anyone else’s. Sequence leads are driven by the sequence builder itself, so you do not build an agent on them. Invoices and knowledge base articles can be acted on by an agent, but an agent is not built on one yet.

Two record types carry a narrower rule: a campaign agent runs only when a campaign finishes sending, and a goal agent only when a goal is achieved or falls at risk.

The actions you have chosen come first, in five groups: Looks up, Drafts, Changes records, Sends and shares and Money and deletes. A line under the heading counts the chosen actions that do something, for example Drafts 1 · Changes records 2. Everything else is under Add an action, searchable and grouped the same way, and Clear beside the heading takes off every action you are able to grant. When setup chose an action, its row can quote the words that asked for it, for example Because: “draft a follow-up to the main contact”.

An agent you build can also use other actions its owner is allowed to use, even when they are not on its list, as long as they fit the records it works on. Only actions under Looks up and Changes records are added this way, and it asks first before any change it makes that way. Actions under Drafts, Sends and shares, and Money and deletes are used only when they are on its list.

Actions under Looks up carry no decision: Looking things up never needs your OK. Every chosen action that changes something carries three positions:

A row you have not decided shows the level’s answer beside the muted word Default, so the next section decides it. A row on the always-asks list shows Always asks, and its Allow cannot be picked. When two answers disagree, the order is fixed: Never wins, then the always-asks list, then Ask first, then Allow, and only then the level. First match wins.

Some changes an agent can only ever propose: tagging a company, setting its status, turning a lead into a customer, putting companies, contacts or a ticket on a new owner, changing a ticket’s status, adding a note only your team can see, recording a promise to pay and holding the reminders on an invoice. Their rows show Always asks. When the agent wants one, it puts a card in front of you and nothing changes until someone approves it. An approved change runs as the teammate the agent runs as, and the card records who approved it.

Most other record changes the assistant can make work the same way for an agent: moving a line on a deal, pausing a campaign or editing a help article waits on a card until someone approves it. Sends and money have one more rule. An email to a customer, a change that starts, rewrites or re-addresses automatic customer email, a payment, a stock ledger entry or an export runs as the teammate the agent runs as, so only that teammate or an admin can approve it, and its card says which mailbox it goes out from, or who it acts as.

One section, two rows, with both lines under each so you can compare them without clicking through. A new agent starts on the first, and setup never moves it higher: only you do. There used to be a third level, Act on routine work, between them. It is now part of Act automatically: an agent that was set to it shows Act automatically and decides the same way.

Always asks, whatever you choose: deleting records · sending anything to a customer · taking or refunding a payment · anything that cannot be undone.

That line sits under both rows at every level, and it is not a setting. Looking things up never waits for you, and a draft is only ever saved as a draft for you to send. For an action that changes something, three things decide it, and it helps to know them:

• The always-asks list also holds any change Laureo rates as high or critical risk, so those wait for a person at every level too.
• An action set to Allow runs without asking even on Ask before acting, because the decision on the action comes before the level. Never always stops it.
• For changes, Act automatically runs any change that is not on the always-asks list without asking, and Ask before acting asks first.

There is no score to tune: for a change, the risk the action carries, the decision you gave it and the level decide, and nothing else.

The Build page keeps a rail that lists the page in its own order: on a wide screen it sits on the left and stays in view, and on a narrow one it sits above the sections. Each stage sums up what is set, and pressing it scrolls to its section.

Everything your agents use counts against your plan’s monthly AI credits, the same AI credits the rest of Laureo’s AI shares. There is nothing to configure and no limit to set: the AI credits come from your plan, and they grow when your plan does. Setup and Try it each say so once: Uses your plan's monthly AI credits.

You are told where you stand in one sentence, in exactly two places: under the header of the Agents home, and at the foot of a test run. Both give a count of AI credits: on the Agents home Used about 8,000 of 10,000 AI credits this month, and on a test run the same count with the date your billing cycle resets, Used about 8,000 of 10,000 AI credits this month, resets Sep 14. The word “about” is always there, because the figure is an estimate and saying so is better than defending a number. On the Agents home the sentence shows the workspace's figure, so it appears there only for people who may see it: people with the billing.view permission, and everyone once a super admin turns on member usage visibility.

When not enough monthly AI credits are left for the next run, your agents pause themselves, and the Agents home says so in front of the count: Paused: not enough AI credits left this month. The count stays the AI credits actually used, and someone who may not see the workspace's figure reads the pause alone. They resume on their own when the cycle resets, and Extra AI Credits start them again before then. If the figure cannot be read at all, the sentence says so rather than showing a confident zero: AI credits usage is unavailable right now.

Content read from records, files, emails, triggers, and earlier tool results is treated as untrusted data. That safety state remains attached through later hops, retries, worker recovery, compaction, and approval continuation. If Laureo cannot determine whether the run is tainted, it treats the run as tainted.

On the Build tab, Try it is the footer control just before Publish, and the test opens directly under the footer. On the Runs and Settings tabs it sits in the page header. Wherever it opens, it says the same two things: Runs on your real data. Won't send or change anything. Uses your plan's monthly AI credits. Beside the record picker it names what it will run: Runs the saved draft. on a draft agent and on a live agent whose edits are waiting, and Runs the published version. otherwise. A new agent can be tried once its first change has saved.

While a test runs from the Build tab, the rail down the side lights up each stage as the test reaches it, and it keeps one stage more than the Build page has, at the end:

Changes save as you go. Publish to make them live. The footer at the end of the page says where things stand on its left: Saving… while a change is on its way, Saved once it has landed, and Didn't save. with a Retry link when it did not. Moving between Build, Runs and Settings never loses a change.

On the right of the footer, Publish comes last: Publish changes on an active agent, and Publish & activate on a draft or a paused one. Every publish keeps the previous setup as a version you can restore. If you move to another page in Laureo while a change is still saving, or after one did not save, the page tries to save it first and, if it has not landed after a moment, asks, under Still saving your last change or Your last change didn't save: Save, Discard or Keep editing.

A live agent’s Build tab opens on the published version, read only, unless edits are already waiting, in which case it opens on them. The footer says This is the published version. Editing starts a draft; the live agent keeps running. Choose Edit as draft in the footer to start changing it. Edits on a live agent's Build tab wait until you publish. A new name and Settings changes save straight away. While edits wait, the footer says Unpublished changes. The live agent keeps running the published version. Try it tests the waiting edits, and Discard changes throws them away after one confirmation, Discard unpublished changes?

Publishing while a run is actually in flight asks you once which you want: Pause and save or Save and keep running. On every other publish there is no confirmation at all. Records already mid-run finish on the version they started, and new runs use the new one.

The Settings tab on an agent holds its identity, its notifications and its archive control. Under Notifications there is one switch, Tell me when this agent keeps failing, on for every agent that has never been touched. Off leaves this agent out of the repeated failure notifications. Every run is still in its history. With it off, the tab reads Repeated failure notifications are off for this agent. It changes nothing about any other agent.