Agent automations
An agent automation gives an agent a task and a schedule, for example "Every Monday at 10:00, sum up last week's new orders by client" or "Every day at 07:00, list the invoices that fall due this week". At each run the agent works on the task the way it answers in Chat, with the same tools and the same approvals, and its answer is saved in a conversation of its own. The agent can be one of your AI employees (see AI team) or Quanty itself. The runs happen on your organization's company computer.
Agent automations are in beta: what they can do may still change.
Before you start
- Plan. Agent automations need the Gold plan or higher (see Plans and billing). On a lower plan the page says "Agents on a schedule need Gold". There is no limit on how many an organization has.
- Who changes what. Members and admins create agent automations. Only the person who set one up, or an organization admin, can change, run, pause or delete it, or decide on a step that waits for approval. Everyone else sees it and changes nothing.
- Whose access. A run works as the person who set up the automation: the agent reads and changes only what that person may, and only inside the agent's own rules: its step of independence, its areas and its access (see AI team). You can schedule only an agent you can use. If the owner of an automation can no longer use its agent, for example because it stopped being shared with them, the next run fails and the automation pauses. An admin who rewrites the task or changes the agent of someone else's automation becomes its owner: from then on its runs work with the admin's access. An admin can do that, or press Run now, only when the automation's agent is one the admin can use; renaming, pausing and rescheduling are always open to admins.
- Who sees what. The task, the summary of each run and the details of a step that waits for approval are shown only to the person who set the automation up and to organization admins. Other members see that the automation exists: its name, agent, schedule, status, next run and the times of its runs. The conversation with the answers belongs to the owner alone; admins cannot open it either. When the agent is someone's private AI employee, people who cannot use it see only that a private agent does the work, without its name or face; the same goes for the Schedule and Computer tabs.
- Credits. Every run is AI work and spends the organization's credits, like an answer in Chat (see What it costs).
The Automations page
Open Automations in the sidebar: "Agents that do a task on a schedule, and the automations of your tables." The page has four tabs:
| Tab | What it shows |
|---|---|
| Automations | The list below, and Your AI team: the employees you can ask right now. |
| Schedule | "What runs on its own in the next 7 days": agent automations, scheduled table automations, table triggers, TMS syncs and the messages your employees may write first (see The Schedule tab). |
| Runs | What ran and how it ended (see Runs). |
| Computer | The company computer: whether it is asleep or awake, what is working now, quiet hours, and the switch that stops everything agents do on their own (see Company computer). |
Automations lists the agent automations of your organization together with the automations of your projects' tables, so both kinds are in one place. Table automations are still built and changed inside their project (see Automations).
Each agent automation is one row:
| Column | What it shows |
|---|---|
| Automation | The orb of the agent that does the work, the name of the automation, and the agent's name under it. |
| Schedule | When it runs, in words, for example "Every Monday at 10:00". |
| Next run | When it runs next, for example "in 3 days". |
| Status | Active, or why it is paused (see When an automation pauses). |
| ⋯ | Edit, Run now, Pause or Resume, Open in chat, Runs, and deleting it. |
Create an agent automation
- Click New automation, then Agent automation ("An agent works on a schedule").
- Under Who does it, pick the agent, or Quanty ("Quanty itself, with every tool of the chat").
- Give it a Name, for example "Monday summary".
- Under What to do, write the task in plain words (up to 8,000 characters), the way you would ask in Chat: what to look at, what to do and what the answer should hold. For example "Check the orders delivered yesterday that have no CMR yet, and list them by client."
- Under When, set the schedule: every day, on chosen weekdays or every few hours, with the time and the time zone. These are the same choices as a table automation's schedule (see Triggers and conditions).
- Click Create automation. While it is active, the list shows when it runs next.
"Each run uses AI credits, like a chat message. The answers are saved in the automation's own chat."
Start from a template
Templates, under the list (or From a template in New automation), offer ready-made tasks grouped by purpose: All, Communication, Productivity, Finance, Research, Sales, Operations and Other. A template fills in the task and a schedule. Pick the agent, change what you like, and save.
A specialist you hire can bring its own suggestions: tick them under Also set up while hiring (see Suggested automations).
Ask in Chat
You can also ask the assistant to set up an agent automation, for example "Every Friday at 15:00, sum up this week's sales for me". It can also pause, resume, rename or reschedule one, change its task, and list the ones you have. A task set up from chat can be up to 1,000 characters. Before it creates or changes anything, it shows what it is about to do and waits for your approval (see AI assistant).
The automation's conversation
Each agent automation has its own conversation with its agent, named after the automation, in the Chat of the person who set it up. Every run adds its answer there, so that person can read what the agent found or did and look back at earlier runs. Open in chat opens it. Nobody else can open this conversation, admins included. When the agent changes, or an admin becomes the owner, the next runs start a new conversation.
Admins read the short summary of each run on Runs; other members see only when it ran and how it ended.
Runs
Runs lists the runs of your organization's automations that you can see, newest first: the runs of agent automations and of table automations together. Show runs filters them: All, Waiting or Failed. A run of an agent automation shows its agent and automation, when it ran, whether the schedule or a person started it, the credits it spent and, when it did not finish, why. The owner and admins also see a short summary of the answer.
| Status | What it means |
|---|---|
| Queued | Waiting to start. |
| Running | The agent is working on the task. |
| Done | The answer is ready, in the automation's conversation. |
| Failed | The run stopped with an error, which it shows, for example when it took too long, Quanty was being updated, or the agent's settings changed since the owner last confirmed the automation (see When an automation pauses). |
| Did not run | The run was skipped, with the reason: the previous run had not finished yet, an AI spending limit was reached, the automation was paused, the organization is read-only, "Skipped: the company computer was asleep." or "Skipped: it was due during quiet hours." The last two come with Open the computer (see What is skipped). A skipped run is not made up later. |
| Waiting | The agent reached a step that needs a decision (see below). |
| Cancelled | The waiting step was rejected, nobody decided within 24 hours, the automation's task, agent or owner changed while it waited, or the automation was deleted. |
Runs that wait for a decision
In Chat, some steps wait for your approval before they happen, for example sending something outside Quanty or deleting (see Confirmations). An agent automation keeps the same rule. When a run reaches such a step, it stops and waits:
- Runs shows it under Waiting for a decision. The owner and admins see a card that says exactly what the agent wants to do, the same card you approve in Chat; other members see only the name of the step;
- the person who set up the automation gets a notification;
- that person, or an organization admin, approves or rejects it.
Once approved ("Approved. The run goes on."), the run goes on where it stopped. Once rejected ("Rejected. The run finishes without it."), the step does not happen and the agent does not continue: the run ends as Cancelled, with a short closing line as its summary that says the step was declined and not done. The same line appears in the automation's conversation. If nobody decides within 24 hours, the run is cancelled. While a run waits, the automation's next scheduled runs do not run.
Approvals still apply
An automation runs while nobody is watching, so every step that needs your approval in Chat waits for a person here too. Other changes, such as editing records, go ahead as they would in Chat. An employee on the step Acts with approval asks before every change, so its run waits at the first one.
What it costs
There is no separate fee for agent automations and no limit on how many you have. Each run is paid in credits from the organization's balance, like an answer in Chat: the model's work at the usual rates, plus anything else the agent uses that costs credits, such as a web search (see Credits and usage). It counts toward the AI spending limit of the person who set it up, if an admin set one. Runs shows what each run spent.
Quanty never repeats a run by itself, so a run is never paid for twice. A run that is cut short, for example while Quanty is being updated, shows as Failed, and the next run on the schedule goes ahead as usual.
When an automation pauses
An automation that has paused stays paused until its owner or an admin resumes it; it never starts again by itself.
| Status | Why | What to do |
|---|---|---|
| Paused | Someone paused it. | Resume it. |
| Out of AI credits | The organization ran out of credits. | Top up or wait for the new month, then Resume it. |
| Needs Gold | The organization is on a plan below Gold. | Move to Gold or higher, then Resume it. |
| Paused after errors | The agent's settings, such as its instructions, model, skills, access, hand-off person, independence, areas, knowledge or daily limits, or the owner of the agent or of the automation changed after the owner confirmed the automation. This also happens when owners edit their own agent. The owner gets a notification. | The owner checks the automation, then Resume or Run now. |
| Paused after errors | The person who set it up has left the organization or is now a viewer, its agent was deleted or paused, or the owner of the automation can no longer use the agent. | An admin picks another agent or becomes the owner, then resumes it. |
While the organization is read-only, no automation runs (see Plans and billing). While the company computer is in the mode Asleep, or during its quiet hours, scheduled runs are skipped and the automation stays Active; Run now works during quiet hours but not while the computer was put to sleep (see Company computer).
Good to know
- What an agent can do in a run. The same as in Chat: work with tables and records, the calendar, files, the TMS when your organization has it, and the chat's other tools.
- Table automations stay as they are. Automations that react to new records, forms, e-mails, WhatsApp messages or KSeF invoices are built in their project (see Automations).
- Agents ask each other here too. During a run, the agent can ask another agent a question, as in Chat (see AI team).
- Every run leaves a trail. What the agent did in a run, and what it read, is in its Activity (see Activity).
- Deleting an automation stops its runs. Its answers so far stay in the chat.
Related pages
- AI team: the employees that do the work.
- Company computer: where the runs happen, quiet hours and the Schedule tab.
- Automations: automations inside a project.
- AI assistant: confirmations and what the assistant can do.
- Credits and usage: credits and the Usage page.
- Plans and billing: what Gold and higher include.