Run an agent from chat and WhatsApp
Give an agent a Chat Trigger and anyone with access can run it by asking for it on the Chat page, on WhatsApp or in Microsoft Teams. This guide covers setting up the trigger, what people see while the agent runs, and how the result comes back.
Checked against the app on
On this page13 sections
- 1What a Chat Trigger does
- 2Before you start
- 3Add a Chat Trigger
- 4Name it and say when to use it
- 5Ask for the details it needs
- 6Ask for it in chat
- 7Follow the run step by step
- 8Get the result back
- 9When a run takes longer
- 10Runs that stop for an approval
- 11Run it from WhatsApp or Teams
- 12Who can run it, and whose accounts it uses
- 13Limits and good to know
What a Chat Trigger does
A Chat Trigger lets people run an agent by asking for it in plain language, instead of waiting for a schedule or an email. Once the agent is published, anyone with access can type something like "Chase the overdue invoices for Northgate Housing" on the Chat page. Centrus recognises the request, picks out the details the agent needs, runs the agent, and answers in the same conversation with what the agent found or did.
Start conversations with Centrus
Your personal chat history, excluding agent-generated threads.





John Martin
last message sent: 1 minute ago
Chase the overdue invoices for Northgate Housing
I ran Chase overdue invoices for Northgate Housing. Three invoices are overdue, and a reminder has gone to their accounts contact for each one:
| Invoice | Site | Amount | Due |
|---|---|---|---|
| INV-20418 | Riverside House | £1,840.00 | 12/09/2026 |
| INV-20433 | Harbour View Apartments | £620.00 | 19/09/2026 |
| INV-20457 | Riverside House | £2,115.50 | 26/09/2026 |
A fourth invoice, INV-20462, is not due until 16/10/2026, so it was left out.
Centrus may make mistakes. Check important info.
- 1Your request, written as you would ask a colleague.
- 2The Worked panel, with the agent's run card inside it.
- 3Centrus's answer, written from what the agent returned.
- 4The message box, for follow-up questions or another request.
The agent itself is built in the Agent Builder like any other, and the Chat Trigger is its first step. An agent has exactly one trigger, so an agent that starts from chat does not also run on a schedule. Choose how an agent starts compares the other triggers. To build or change agents by sending messages, rather than run them, see Build and edit agents from chat and WhatsApp.
Before you start
Three things need to be in place before Centrus will run an agent from chat:
- Agent Builder chat tools are switched on for your organisation. An admin turns on Enable Agent Builder chat tools in Settings > Organisation. It is off by default.
- The agent is published. Drafts and paused agents are not offered to chat, and chat always runs the version you last published, not the changes you are still working on.
- You have Use or Edit access to the agent. Whoever created it always has it, and everyone else needs it granted in Manage access. View access is not enough, and being an admin does not give access on its own. See Share an agent.
Enable Agent Builder chat tools
Lets workflows built in the Agent Builder with a chat trigger be called directly from this chat as tools, and shows their run history to the AI when asked. Off by default.
If any of the three is missing, Centrus does not know the agent exists. It will not show an error; it simply answers the message as an ordinary question. When someone says chat ignored their agent, check these three things first.
Add a Chat Trigger
- 1Open the agent in the Agent Builder.
- 2In the Nodes panel on the left, find Chat Trigger under Trigger, and drag it onto the canvas or click it. If the agent already has another trigger, delete that one, because an agent can only have one.
- 3Connect the Chat Trigger to the first step of the work, such as a BigChange REST step that looks up invoices.
- 4Click the Chat Trigger to open its details, and fill in Set up and Parameters as described below.
- 5Publish the agent. It is available in chat as soon as publishing finishes.
Until the trigger has a tool name, its card on the canvas reads Tool name not set. Once it has one, the card shows Tool: followed by the name, so you can see at a glance what chat knows the agent as.
Chat Trigger
Tool: chase_overdue_invoices
Fetch unpaid invoices
List invoices
Overdue invoices only
Keep ALL of 2 rule(s)
Write reminder emails
GPT-5.6 Terra
Send reminders
1 action enabled
Set the trigger up on a computer. On a phone or in a narrow window, the builder hides the Nodes panel and the details panel.
Name it and say when to use it
The Set up box in the Chat Trigger's details holds the two fields that decide whether Centrus runs your agent.
Chat Trigger
Tool: chase_overdue_invoices
Fetch unpaid invoices
List invoices
Overdue invoices only
Keep ALL of 2 rule(s)
Write reminder emails
GPT-5.6 Terra
Send reminders
1 action enabled
- 1Tool name: the agent's name as chat sees it.
- 2Description: what the agent does and when Centrus should run it.
- Tool name is the agent's name as chat sees it, for example chase_overdue_invoices. Use lower case letters, digits and underscores, start with a letter, and keep it between 3 and 64 characters. The run card in chat turns it back into words, so chase_overdue_invoices appears as Chase overdue invoices.
- Description tells Centrus what the agent does and when to use it. It is the only thing Centrus reads when deciding whether a message is a request for this agent, and the people using chat never see it.
Write the description as an instruction. Say what the agent does, then the kinds of request that should start it, for example: "Finds a customer's overdue invoices in BigChange and emails a reminder for each one. Use this when someone asks to chase, follow up or send reminders about overdue or unpaid invoices for a named customer." A vague description such as "Invoice helper" leaves Centrus guessing, so the agent either fails to run when it should, or runs when someone only wanted to ask about an invoice.
Each tool name can belong to only one published agent in your organisation, and names that start like Centrus's own tools are blocked, such as outlook_, gmail_, bigchange_, simpro_, xero_ and excel_. Publishing stops with a message if the name is missing, badly formed or already taken, so choose a name that describes the job, such as chase_overdue_invoices rather than outlook_reminders. Duplicating an agent gives the copy a new tool name automatically.
Ask for the details it needs
Parameters are the details Centrus should pick out of a request and hand to the agent, such as a customer, a site or a number of days. Centrus fills them in from what the person typed, so "Chase anything Northgate Housing owes us that is more than 30 days late" becomes a customer of Northgate Housing and a days overdue of 30.
- 1In the Parameters section, click Add parameter.
- 2Click the new row to open it, and enter a Field name, such as customer.
- 3Choose its type: String, Number, Boolean, Array or Object. Most parameters are a String or a Number.
- 4Describe it in the box underneath, for example "The customer's name as it appears in BigChange, such as Northgate Housing." Centrus reads this to work out what to put there.
- 5Tick Required if the agent cannot do its job without it.
Chat Trigger
Tool: chase_overdue_invoices
Fetch unpaid invoices
List invoices
Overdue invoices only
Keep ALL of 2 rule(s)
Write reminder emails
GPT-5.6 Terra
Send reminders
1 action enabled
- 1Add parameter adds a new, empty row.
- 2The Field name and its type.
- 3The description Centrus reads to fill the parameter in.
- 4Required, for details the agent cannot run without.
Later steps use a parameter by its field name, for example {{ trigger.json.customer }} in an Intelligence step's instructions or in a BigChange REST filter. Every parameter needs a name and no two can share one; publishing stops until both are true.
With no parameters, the section reads No parameters: the tool will be called with no arguments. That suits an agent that always does the same thing, such as a job summary for the whole company. Centrus can ask the person for a missing required detail before it runs the agent, so a clear description for each parameter saves a round of questions.
Ask for it in chat
Open Chat, start a new chat or open an existing one, and ask for the work in your own words. You do not need the tool name or a special command, and the chat's tags make no difference. All of these would run the agent above:
- Chase the overdue invoices for Northgate Housing.
- Can you send reminders for anything Northgate Housing owes us that is more than 30 days late?
- Northgate Housing still has not paid. Please chase them.
Centrus starts the agent as soon as it recognises the request. The Working panel at the top of the answer names the agent, followed by the main detail it passed on, for example Working · Chase Overdue Invoices · Northgate Housing, and a blue run card appears inside it.
John Martin
last message sent: less than a minute ago
Chase the overdue invoices for Northgate Housing
If Centrus answers without running the agent, it did not match your message to the agent's description. Ask more directly, for example "Run Chase overdue invoices for Northgate Housing", and let whoever built the agent know so they can make the description clearer.
Follow the run step by step
The run card follows the agent through its steps. While the run is going, it reads Running and the agent's name, with Completing step and a count underneath. When the run ends, the header changes to Completed or Failed, with a line such as 5 of 5 steps completed. Click the card to open the list of steps.
- 1How the run ended, and how many steps finished.
- 2Each step in order, named by its step type.
- 3How long the step took, and its status.
- 4A one-line summary of what the step did. Click a step to read it in full, with any error.
- Completed, in green, means the step finished.
- Running, in blue, means the step is working now. A Browser Use step that is busy on a website also shows as Running.
- Awaiting approval, in amber, means the run has stopped at an Approval step.
- Failed, in red, means the step hit an error. Click it to read the message.
- Skipped and Cancelled appear in grey.
Steps are named by their type rather than the label you gave them on the canvas, so two Intelligence steps both appear as Intelligence. The card checks for updates every one and a half seconds. The same run, with each step's output, is in the agent's Runs panel; see Read the Runs panel.
Get the result back
When the run finishes in time, Centrus reads what the agent returned and writes its answer from it, so the reply reads as a normal chat message rather than raw data. You can ask follow-up questions about the result straight away.
Chase the overdue invoices for Northgate Housing
I ran Chase overdue invoices for Northgate Housing. Three invoices are overdue, and a reminder has gone to their accounts contact for each one:
| Invoice | Site | Amount | Due |
|---|---|---|---|
| INV-20418 | Riverside House | £1,840.00 | 12/09/2026 |
| INV-20433 | Harbour View Apartments | £620.00 | 19/09/2026 |
| INV-20457 | Riverside House | £2,115.50 | 26/09/2026 |
A fourth invoice, INV-20462, is not due until 16/10/2026, so it was left out.
Which contact did the reminders go to?
Only the last step's output comes back to chat. If the agent looks up invoices and then sends emails, Centrus receives what the email step returned. Make sure the last step produces what the person needs to know, for example an Intelligence step that writes a short summary.
Files do not come back as attachments. A PDF made by a Create PDF step stays inside the run, so to hand a document back, end the agent with a step that saves it, such as OneDrive, SharePoint, Excel or Word. Those steps return a link, and Centrus includes it in the answer.
If the run fails, Centrus says so in its answer and passes on the agent's error message. The same happens when your organisation has used its monthly automation credits or already has as many agents running as it is allowed.
When a run takes longer
Centrus waits for the agent for up to the Sync timeout (seconds) set on the Chat Trigger. The default is 180 seconds and the most you can set is 300. If the agent is still going when the time is up, Centrus stops waiting and replies that it is still working. The agent carries on in the background.
John Martin
last message sent: less than a minute ago
Run the daily job summary for Riverside House
The Daily job summary for Riverside House is still running, as it is taking longer than usual. Ask me to check on it in a bit; there is no need to ask for it again from scratch.
On the Chat page, the still-working reply keeps the live run card above it, so you can watch the agent finish. When the run ends, Centrus replaces the still-working reply with the result. The new text appears the next time the chat loads, so refresh the page or reopen the chat to read it.
Run the daily job summary for Riverside House
✅ Finished the workflow you asked me to run in the background.
Inserted 6 blocks into Word document
- Open document
The finished reply starts Finished the workflow you asked me to run in the background, followed by the last step's summary and a link to each document or file it produced. If the run failed instead, it starts The workflow you asked me to run in the background failed, followed by the error.
You can also ask, for example "Has the daily job summary finished?" The Working panel shows Checking automation progress, and Centrus tells you whether the run has finished and what it did last. Do not ask for the agent again while it is running, because that starts a second run.
Raise the timeout for an agent that usually takes three or four minutes, so people get the answer in one go. "Still working" message (optional), under the timeout, is meant to replace the standard still-working reply with your own words, such as how long the agent normally takes. At the moment the field is saved but chat still uses its standard reply, so there is no need to fill it in.
Chat Trigger
Tool: chase_overdue_invoices
Fetch unpaid invoices
List invoices
Overdue invoices only
Keep ALL of 2 rule(s)
Write reminder emails
GPT-5.6 Terra
Send reminders
1 action enabled
Runs that stop for an approval
If the agent has an Approval step, the run stops there until someone approves or rejects it, and chat stops waiting straight away. Centrus replies that the agent is waiting on an approval step and cannot finish automatically from chat, and the run card shows the Approval step as Awaiting approval. The card's header keeps reading Running while the run waits, so look at the step list for the real state.
Chase the overdue invoices for Northgate Housing
Chase overdue invoices is waiting on an approval step, so it cannot finish automatically from chat. The reminders will only go out once someone approves the run.
The approver decides in the agent's Runs panel, or by replying to the approval email if the step sends one. The run then carries on by itself, but its result is not posted back to the chat or to WhatsApp. To see how it ended, ask Centrus whether the agent has finished, or open the run in the Runs panel. Conditions, loops and approvals covers Approval steps in full.
Run it from WhatsApp or Teams
The same agents are available when you message Centrus on WhatsApp or in Microsoft Teams, with the same access rules. Ask in your own words, as you would on the web; the conversation's tag makes no difference.
WhatsApp has no run card. You see Message received, generating response, please wait… and then the answer when the run finishes, which can take up to the sync timeout.
If the run outlasts the timeout, Centrus says it will message you as soon as the agent is done. When the run ends, a separate WhatsApp message arrives starting Finished the task you asked me to run in the background, with any links written out in full.
In Teams, the answer arrives as a normal reply once the run finishes. A run that outlasts the timeout does not message you in Teams when it ends. Its result is added to the conversation's history on the Chat page instead, so ask Centrus in Teams whether it has finished, or read the conversation on the web.
Chat
Centrus09:14
I ran Chase overdue invoices for Northgate Housing. Three invoices are overdue, and a reminder has gone to their accounts contact for each one:
- INV-20418, Riverside House, £1,840.00, due 12/09/2026
- INV-20433, Harbour View Apartments, £620.00, due 19/09/2026
- INV-20457, Riverside House, £2,115.50, due 26/09/2026
Who can run it, and whose accounts it uses
Centrus only offers an agent to people with Use or Edit access to it, so two people sending the same message can get different results: one runs the agent, the other gets an ordinary answer. Access is set in Manage access, which you open from Settings in the builder; see Share an agent.
By default, an agent's Outlook and Gmail steps use the agent's own connection, whoever asks. For an agent that sends email, you can have it send from the account of the person who asked instead. Open Settings in the builder toolbar and, in the Agent settings window, choose Per-user under Integration mode. Only the agent's owner and admins see this option, and it can only be changed for agents that start from chat.
Live and running on its trigger. Switch to Draft to pause it while you make changes.
Control who can view, use, or edit this agent: in the builder and from the Chat page.
Manage accessOutlook (send/draft) and Gmail steps use the credentials of whoever actually triggers this workflow, not this agent’s own connection.
- Agent-level uses the agent's own shared connection for every run.
- Per-user uses the Outlook or Gmail account the person who asked has connected in their own integrations. Only sending emails and creating drafts work in this mode; any other Outlook or Gmail action makes the run fail with a message asking you to switch back.
Other systems, such as BigChange, always use the agent's connection. A change to Integration mode applies to the published agent straight away, without publishing again. To connect your own mailbox, see Microsoft Outlook in chat or Gmail in chat.
Limits and good to know
Chat waits
180 seconds
By default, up to 300. Longer runs finish in the background.
The result
Last step
Only the last step's output comes back. Files come back as links.
Cost
1 credit a run
Plus each step's credits. The Chat Trigger itself is free.
Who can run it
Use or Edit
Published agents only, for people with access to them.
- Credits. The Chat Trigger costs nothing. Each run uses 1 automation credit to start, plus the credits for its steps, within the agent's per-run budget of 250 credits unless it has been changed. The chat answer around it counts as a normal chat message. See Automation credits and usage limits.
- Runs at once. An organisation can have 10 agent runs going at the same time by default. A request beyond that fails, and Centrus passes on the message saying the limit has been reached.
- Testing. Run now, in the agent's Runs window, fills each parameter with a stand-in: sample value for text, 0 for numbers and true for Booleans. Steps that look up a real customer will find nothing, so test from chat once the agent is published, bearing in mind that a chat run is a live run. Test an agent before it goes live covers safer ways to try it first.
- Paused agents. Pausing an agent or switching it back to Draft removes it from chat until it is published again.
More in Agent Builder
- Open Agent Builder
- Build your first agent from a message
- Build and edit agents from chat and WhatsApp
- Find your way round the canvas
- Choose how an agent starts
- Run an agent on a schedule
- Start an agent when an email arrives
- Start an agent from a system alert
- Steps an agent can use
- Conditions, loops and approvals
- Send text messages
- Make outbound calls and answer incoming calls
- Test an agent before it goes live
- Publish, pause and restore versions
- Read the Runs panel
- Share an agent
- Automation credits and usage limits