Automations
Give AI steady work. You hold the keys.
An agent has a role, a schedule and permissions, and it works even when the app is closed. Emails, messages and publications wait for your approval.
Set up an agent: every morning at 7:30 draft replies to emails, send nothing.
- When
- weekdays, 7:30
- Permissions
- Reading mail · AutoDrafts · AutoReminders · AskSending · Never
- Limit
- 60 credits per run
Adding a reminder: “Roof inspection, Thursday 10:00” (deadline from the client's email).
From a sentence to an agent in five steps.
You say once who does what and when. Then the agent works on its own, and you decide only where you want to decide.
- 01Describe the task or pick a template
In the conversation, write what the agent should do and when, in the app or in the browser, e.g. “Set up an agent: check my competitors' news every day and let me know in the morning”. The assistant will pick a ready template or suggest your own agent, and will ask one question about any missing detail.
- 02Set permissions
The assistant will show in the conversation (and the iOS app on the agent's screen) what the agent does on its own and what it prepares for your approval: Auto, Ask or Never. Sending emails, WhatsApp and publishing wait for approval by default, and the agent does not see tools set to Never at all.
- 03Test run
A test run reads real data, but only simulates emails, posts and changes. You see the result and the cost before the agent really starts.
- 04Approve with one tap
A client email, a calendar event, a WhatsApp message: each such action waits in the approvals inbox. You approve exactly what you see.
- 05Check results and history
The result arrives by email, and a task created in a conversation can send it back to that conversation. You can ask the assistant about the last run. The full history with cost and action log is in the iOS app (from version 0.1.15), and we keep runs for 180 days.
Ready-made agent templates. You fill in a field or two.
A template is an agent with a role, permissions and a task with a schedule. You pick it in the app (Automations screen, ⋯ menu → Templates) or ask for one in a conversation, e.g. “Set up an agent: industry press review”, and change the time and details later.
Most templates work right away. Two Facebook templates require Meta's consent, and Shopify and Google Ads are coming soon.
Mail and calendar
What needs a reply, invoices and payments, reply drafts in Gmail. It sends nothing.
A day plan with free slots, and for each meeting the latest email threads, the company in two sentences and three points to raise.
The coming week day by day: meetings, conflicts, free slots and reminders to prepare.
Sales and finance
Rates the enquiry (urgent, regular, spam), prepares a reply for your approval and sends you a push with a summary.
Finds customers who asked for an offer and got no reply, and prepares a follow-up. Sent after your approval.
Checks receivables and prepares reminders graded by how overdue they are. Sends only after your approval.
Marketing and shop
Reads new comments, prepares replies to questions about the offer and waits for approval. Complaints are passed on to you.
Posts with the most reactions, unanswered questions and three post ideas for the new week.
Cost, clicks, conversions and CPA per campaign, expensive keywords without conversions and three recommendations.
Revenue, orders, average basket, week-on-week change, top 5 products and returns.
Knowledge and competition
5-8 most important items from your industry in the last day, each with one sentence “what it means for my business”.
New products, prices, promotions and announcements from competitors' sites. The three most important changes at the top.
Code and IT team
What is waiting for review, red CI, merged since yesterday and PRs open for more than 5 days.
New issues with a summary, type and duplicates. Labels are assigned after your approval.
Your own agent in a ready role.
When no template fits, the agent is built from scratch. The role gives it ready instructions and sensible permissions, and an example shows what it is good for.
Mail and calendar
Inbox, meetings, remindersEvery day at 7:30: what in the inbox needs a reaction today, with reply drafts.
Marketing and social
Posts, comments, statistics, adsOn Monday: three proposals for the week's posts with copy and graphic ideas.
Sales and offers
Offers, leads, customer contactAfter a form submission: a reply with an offer, sent after your approval.
Researcher
Searches, reads, summarizesEvery week: industry news with sources.
Data and reports
Spreadsheets, figures, chartsOn Monday at 8:00: a week summary from project files with a chart.
Code and servers
Scripts, repositories, running codeIn the morning: new issues and failed builds on GitHub.
Assistant
General office tasksOn Friday: a list of tasks for next week from meeting notes.
AI automation where you have the last word.
Each agent skill has one of three modes. Auto: does it on its own, and the action goes into the history. Ask: prepares the action and waits for your approval. Never: no access.
| Skill | Default mode | Why |
|---|---|---|
| Read the Gmail inbox | Auto | Reading sends nothing and changes nothing. |
| Send emails on my behalf | Ask | A message to another person goes out only after your approval. |
| Add and change events | Ask | A change in the calendar also affects invited people. |
| Send WhatsApp messages | Ask | The customer gets the message only after approval. |
| Publish sites | Ask | A publication is visible to everyone. |
| Create further automated tasks | Never | The agent does not plan new work for itself. |
| Meta campaigns and ad budget | Unavailable | The Meta Ads connector requires Meta's consent. Until Meta grants it, the agent does not have these skills. |
No high-risk skill is set to Auto by default. When you set up an agent in a conversation, the assistant switches a skill to Auto only after your explicit consent in that conversation. Risky approvals are confirmed with Face ID, and a request without a response expires after 7 days.
You pay for work, not for an agent subscription.
A run is billed in credits, like work in the chat. Each has a starting cost of 20 credits (55 credits for tasks on a Computer machine) and a credit limit: 60 credits by default, 500 credits at most. When the agent uses it up, it stops and describes what it managed to do.
- Agents on every plan, including Free. From 1 agent and 2 tasks on Free up to 50 agents and 100 tasks on Max.
- The cost of each run is shown in the task history, and a monthly estimate before you create the task.
- Never more than you have on your account. When credits run out, the run is skipped.
Rates: plan pricing and pricing in credits.
| Plan | Agents | Active tasks | Most often |
|---|---|---|---|
| Free | 1 | 2 | every 24 h |
| Base | 3 | 5 | every 12 h |
| Plus | 10 | 15 | every 6 h |
| Pro | 25 | 40 | every 3 h |
| Max | 50 | 100 | every 1 h |
| Enterprise | no limit | no limit | every 1 h |
On every plan: a credit limit per run of up to 500 credits and 7 days to decide on an approval. Webhooks from the Base plan, daily runs and other limits are in the limits table.
Mail, calendar, GitHub and WhatsApp.
You connect each service once in Connectors, on your own account, and you can disconnect it at any time. When a template or permission requires a connection, the app will show a “Connect” button.
Gmailreading, drafts, sending after approval
Google Calendarday plan, events, reminders
GitHubpull requests, issues, builds
WhatsAppbot on the DomaAI number, own number coming soon
Facebook and Instagramrequires Meta's consentMeta Adsrequires Meta's consentAutomation agents vs Multi-Agent.
Two kinds of agents on one account. They differ in who starts them and where they deliver the result.
| Feature | Automation agents | Multi-Agent |
|---|---|---|
| Who starts it | A schedule, a webhook or “Run now” | You, with a message with Multi-Agent turned on |
| When it works | On a schedule, also when the app is closed | Right away, while you wait for the answer |
| Where the result is | In a notification, email or conversation, with run history | In the conversation |
| Cost | 20 credits to start (55 credits on a Computer machine), run limit 60 credits by default | From 5 credits per run plus model tokens |
Questions about agents
The full list of answers is in the help center.
How do I create my first agent?
The simplest way is in a conversation: write what the agent should do and when, e.g. “Set up an agent: every Monday at 8 send me what's new with my three competitors”. The assistant will pick a template or suggest your own agent, ask about any missing details and set it up. This works in the app and at domaai.pl/czat. In the iOS app (from version 0.1.15) there is also a separate Automations screen with tasks, agents, approvals and templates: you open it from the “+” menu with the tile “Agents and automations”.
Can I test an agent before it starts?
Yes. A test run reads real data (mail, calendar, websites), but only simulates emails, posts and other changes. You see the result and the cost before the task starts on its schedule.
How do approvals work?
You get a notification that the agent is asking for approval. In the approvals inbox or on the card in the chat you see a preview of the action and choose Approve or Reject. You approve exactly what you see, once. Risky approvals are confirmed with Face ID, and a request without a response expires after 7 days.
What is a webhook?
An address to which a form on your site or another service sends data (a POST with JSON). Each call starts a task (and counts toward the plan's daily run limit), and the data goes into the agent's instructions. The iOS app shows the address in task details (from version 0.1.15), and in the browser, the “Machines” tile window in the side panel, under the task in the “Routines” section. Treat it like a password.
How much does a run cost?
As much as the agent's work: model messages and tools are billed in credits as in the chat. A run has a starting cost of 20 credits (55 credits for tasks on a Computer machine), and each agent has a limit per run (60 credits by default, 500 at most). When it runs out, the agent stops and describes what it managed to do. You see the cost in the task history.
How do I stop an agent?
Ask in a conversation, e.g. “pause the morning mail review”, or turn the task off with the switch: in the iOS app on the Automations screen (Tasks tab), in the browser in the side panel: the “Machines” tile, “Routines” section (switch next to the task). In the iOS app (from version 0.1.15) you can also turn off the whole agent with the “Agent active” switch, and pause all tasks at once with the “Pause all tasks” command in the ⋯ menu.
Can I manage agents in the browser?
Yes. On domaai.pl/czat you create an agent in a conversation, decide on approvals (Agent approvals in the side panel, or the link in the email with the result) and turn tasks on or off with the switch in the “Routines” section of the window that opens from the “Machines” tile in the side panel. Detailed editing of the agent (role, instructions, permissions, limit) is in the app on iPhone and iPad from version 0.1.15.
Can I deploy agents in a company?
Yes. The Enterprise plan has individual pricing and no limit on agents, tasks or webhooks (rhythm down to every hour). A VAT invoice is issued automatically after payment on the site. For deployment, ask at biznes@madd.im.
Start with a single task.
Write, for example, “Set up an agent: every Monday at 8 make a report on what's new in my industry”. The agent is created in the conversation, and approvals stay with you.
