Agents AIAI feature

Some work is predictable enough to describe once. When an offer is signed, someone should open the installation file. When an invoice goes past due, someone should chase it. An agent does that work. You give the agent a standing goal, then add jobs. Each job is one trigger plus an instruction in your own words. Agents use the same AI assistant as the in-app chat, with the same tools, but they run without anyone watching.

Before you start

• Agents must be switched on for your company. If you don't see Agents in the navigation, ask your Reonic account manager. The page is labelled Beta.
• You need to be an admin. Only admins can see, create, edit and run agents.
• Decide what the agent is for before you create it. Its overall goal shapes every run.

Create an agent

1. Open Agents. On your first visit a welcome window introduces agents. Click Got it to close it.
2. Click New Agent. Reonic creates an agent named New Agent and opens its page.
3. Under What should the agent be called?, enter a name that tells you and other admins what the agent does.
4. Optionally, add an avatar. Click the dice to generate one with AI from the agent's name, goal and job instructions, so fill those in first. Each click gives a different image in your company's colours. To use your own image, click the pen and upload one. The avatar is applied when you save.
5. Fill in Overall goal: the standing instructions, such as tone, priorities, what to always check and what never to touch. Reonic adds the overall goal above the instruction of every job this agent runs, so put cross-cutting rules here and job specifics in the job.
6. Set Agent Status to Working. While an agent is On vacation, event triggers and schedules skip it, but you can still test it.
7. Leave Automatically run actions that need confirmation off unless you trust the agent to act alone. While the setting is off, the agent pauses before actions such as deleting, archiving, changing a status or sending a signature request, until an admin confirms.
8. Click Save.

The agent has no jobs yet, so it does nothing. One agent can hold many jobs, so you don't need a separate agent per trigger.

You can also click Suggest agents on the Agents page. Reonic proposes ready-made agents based on your company's data, and each card you select creates one agent with one job and a description. Suggestions work best once your account has some usage.

Add a job that reacts to an event

A job fires on one of these events:

• On offer signed: an offer is signed, which also marks the project as won. Residential only.
• On signature requested: a signature is requested, before the offer is signed. Residential only.
• On project status changed: a project moves to a new Kanban status.
• On project marked as won: the project is marked Won. This is a separate trigger from On offer signed.
• On project marked as lost: the project is marked Lost. The agent receives the lost reason, so your instruction can react to why the project was lost.
• On checklist completed: a checklist segment is locked for editing.
• On request created: a new project is created, whether manually, through AI lead import or through public signup.
• On invoice overdue: an unpaid invoice becomes overdue. Reonic checks hourly, and the job fires once per invoice. Invoices whose payment status was changed by hand are excluded.
• On appointment created: an appointment is created in a Reonic team or user calendar, with or without a linked project. Appointments in connected Google or Microsoft calendars don't trigger the job.

1. Open the agent from the Agents list.
2. In the Jobs section, click Add job. Reonic creates an empty, inactive job and opens it.
3. Optionally, enter a Job name. Without one, the job shows as "Job N".
4. Under When should the job trigger?, pick the event. Most options show a help line describing exactly when they fire.
5. Optionally, narrow the job. On project status changed can fire only when a project lands in one status, On checklist completed only for one checklist template, and On project marked as lost only for one lost reason. Leave the field on its "any" option to fire every time.
6. Write the Instruction: what the agent should do when the trigger fires. The instruction is required.
7. Set Status to Active.
8. Click Save. If the job is still inactive, Reonic asks Activate job?. Choose Activate to let the job run, or Keep inactive.

All other triggers fire on every matching event. Put extra conditions, such as "only for projects over a certain value", into the instruction. The agent receives the event details and does nothing when the condition isn't met.

When a job fires on a project, the agent is told to change only that project. The agent can still read other projects.

One job has exactly one trigger. To react to several events, add several jobs to the same agent. A job reacts to events from the moment it's active. Earlier events aren't replayed.

Add a job that runs on a schedule

1. In the agent's Jobs section, click Add job.
2. Under When should the job trigger?, choose At a certain time. The schedule starts as daily at 9 UTC.
3. Pick the Cycle: Daily, Weekly, Bi-weekly or Monthly.
4. For weekly and bi-weekly, choose the Day of week. For monthly, choose the Day of month: Start, Middle or End. End is always the last day of the month, including in short months.
5. Set the Hour of day. Each hour is shown in UTC, with your browser's local time beside it.
6. Write the Instruction.
7. Click Save, then choose Activate when asked.

The schedule runs in UTC, in whole hours. A 09:00 local run is a different UTC hour, and it shifts by an hour when the clocks change. The job starts at a fixed minute within the selected hour, not exactly on the hour. If a slot is missed, that run is skipped rather than caught up later.

A scheduled run has no project attached, so the instruction has to say what to look at. Bi-weekly fires on the chosen weekday in even ISO calendar weeks, not two weeks from when you saved the job.

Test a job

1. Open the job and click Try. Try becomes available once the instruction is saved.
2. Pick a project. Scheduled jobs skip this step.
3. Click Explain what will happen. The agent describes the changes it would make and names anything ambiguous in your instruction.
4. When you're satisfied, click Run now.
5. Confirm the prompt. This run makes real changes.

Test runs act as the agent, not as you. Each Run now really acts, so repeated tests repeat their side effects, such as two follow-up tasks. Explain instructs the agent to make no changes. For your first tries, use a test project.

Approve a paused action

When a run reaches an action that needs confirmation and automatic confirmation is off, the run pauses with Permission needed. Your company's admins get an in-app notification.

1. Open the run from the runs table, the Permission needed tab on the home page, or the notification.
2. Confirm or decline the action, in the run view or directly from the notification.

The first admin to answer decides. A confirmed run continues as the agent. A pending request expires if nobody answers for too long.

Review an agent's runs

The agent's page shows Past runs across all its jobs, and each job page shows that job's last 50 runs. Each row shows one status:

• Queued: waiting to start, because your company already has the maximum number of triggered runs in progress.
• Running: still in progress.
• Permission needed: paused, waiting for confirmation.
• Stopped: someone stopped the run.
• Success: the goal was achieved.
• Partial: part of the goal was done, with a percentage.
• Failed: the goal couldn't be achieved. This includes runs that timed out, crashed or were stopped by a built-in guard.
• Skipped: the trigger didn't really apply, so there was nothing to do.
• Unknown: the run finished without reporting an outcome.

The agent reports the outcome itself, so a step that failed and then succeeded on retry doesn't count against the run.

The Type column shows how a run started: Triggered for a real run, Manual for Run now, and Preview for Explain. Triggered runs carry a depth badge. A run started by an event or schedule shows 1, and a run set off by another agent's run shows a higher number.

Jobs that exist to tell you something rather than change something, such as a digest or an analysis, produce Report runs. They show Success in the runs table and also appear on the home page for you to read.

Each row has these actions:

• Open shows the run's conversation beside the table.
• Stop ends a queued, running or paused run. A running run stops after its current step. Actions waiting for approval are cancelled, and changes already made stay.
• Retry starts a fresh run of any finished run, with the original trigger. A retried Explain run is still a preview.
• Help me fix this, on a run that ended Failed, Partial or Unknown, copies the run into a new private AI chat that diagnoses what went wrong and proposes a change. Edit the job with the fix you agree on.

To pause one job, set its Status to Inactive. To pause all of an agent's jobs, set Agent Status to On vacation and save. Neither deletes anything. Queued runs of the paused job or agent are stopped instead of started.

Check the home page

The Reonic Portal home page has an Agents overview that collects the runs worth looking at across all your agents, in the tabs All, Report, Permission needed and Failed. Each tab shows up to 5 runs, unseen ones first. Click a row to open the run.

You also get an in-app notification when a report is ready, a run needs permission, or a run fails. These notifications go to every admin, whatever their other notification preferences.

Things to know

• An agent with no jobs never fires. The agent holds its name, overall goal, status and confirmation setting. The triggers live on its jobs.
• Automatic confirmation also covers your tests. Once Automatically run actions that need confirmation is on, expect Run now to carry out those actions without pausing.
• Runaway protection is built in. Runs stop after about 10 minutes or 100 tool calls, and a chain of agents setting each other off is capped at three deep. These limits always apply.
• Busy periods queue runs. Up to 10 triggered and scheduled runs per company run at the same time. Further runs show Queued and start, oldest first, as slots free up. They start later and are not dropped. Explain and Run now tests, retries and confirmed actions don't count towards this limit.
• Quick repeats run separately. If the same event fires a job twice in quick succession, you get two separate runs.
• Deleting an agent deletes its jobs with it, in one step.
• Each run uses AI. Keep an eye on frequent schedules that do a lot. Switching a job to Inactive stops it.

Last updated on November 9, 2026