Help Centerchevron_rightGenaya AIchevron_rightAgentschevron_rightAgents that work for you
smart_toy

Agents that work for you

Genaya AI / Agentsschedule9 min readmenu_bookArticle 2 of 7 in Agents
open_in_newOpen Agents in Genayaarrow_outward

You never fill a form to make an agent. Say what it should do and Genaya AI fills in the card: a name, the purpose in your words, the schedule in your organization's time zone or the event that starts it, the actions it may use, a starter text or email, how much it may do on its own and its limits. Nothing turns on until you press Turn on.

Four ways to make one

Say it
Type "make an agent that ..." or a recurring sentence - "every morning tell me who did not pay", "whenever a call is missed, text them back" - in any Genaya AI chat, on the web or on the phone. The card arrives already filled in.
The /agent command
Type /agent and the job: /agent remind clients about unpaid invoices every Monday at 9. The composer arms the agent action and the chat drafts the same card.
Make this an agent
When an answer describes a recurring job - "every morning I can text yesterday's no-shows and ask them to reschedule" - a chip appears under it. Press it and the chat drafts that job as an agent; when your own message was the recurring sentence, your words become the brief.
The gallery
Genaya AI > Agents opens the gallery: a composer at the top, three example pills and eight ready-made agents with a picture each - Lead follow-up, Missed-call text-back, Review request, Collections, Morning brief, Follow-ups, Appointment coordinator and Inbox triage. Turn on lands on the same filled-in card the words would, never an empty form and never an automation.

The Agents page is a gallery, on the web and on the phone. It reads top to bottom:

  1. 1
    The composer at the top

    Type what the agent should do - "text every new lead within 5 minutes" - and send it. A new chat opens with the sentence, and Genaya AI answers with the agent card already filled in from your words. Nothing turns on until you confirm it there.

  2. 2
    Three example pills

    "Text every new lead within 5 minutes", "Every morning tell me who did not pay" and "Ask for a review after a job is done" sit under the composer. Press one and it is filled in and sent in the same press.

  3. 3
    Prebuilt, My agents, Approvals

    Prebuilt shows the eight ready-made agents. My agents shows the ones you turned on, with a count of how many need you. Approvals opens the inbox of cards waiting for your tap.

  4. 4
    One picture per agent

    Every prebuilt agent has its own picture on a soft tint, its name, one line about the job and Details for when it runs, the number it shows, who it is for, what it never does without asking and what it replaces. A card that creates an automation on its first run says so before you turn it on.

  5. 5
    Turn on

    Press Turn on and the chat opens with the same filled-in card the composer would have drafted for that agent - never an empty form and never an automation. An agent that is already on shows its status word (Running, Paused or Needs you), a quiet line with its last run and its number, and its menu offers Pause, Run now, Rename and Delete.

  6. 6
    The Custom agent card

    The last card is Custom agent. Press Describe it and the cursor lands in the composer at the top, ready for your own sentence.

  7. 7
    My agents, sorted by who needs you

    The agents you run appear as the same cards, the ones waiting on you first, each with its last run and its number. Press a card to open the agent and read its transcript.

What the morning brief and the Follow-ups agent look at

Both run every morning at your brief time - 7:30 AM in your organization's time zone by default; change it on the agent's Brief time row - and read your real records first, with no guessing:

  • Yesterday's appointments: done, no-show (the time passed without Done or Canceled) and canceled
  • Today's appointments (the Morning brief counts them; the Follow-ups agent looks only at what needs a reply)
  • Missed inbound calls that never got a text back
  • New leads from the last 2 days nobody has called, texted or emailed
  • Invoices past their due date with a balance still open
  • Estimates still waiting 3 days or more after they went out
  • Jobs marked Done whose client has not been asked for a review

Then it writes one short brief in your own words - "Yesterday you had 4 appointments: 3 done, 1 no-show. Want me to send the 3 done ones a thank-you and reschedule the no-show?" - with up to 6 proposals, each a card. Yes, send confirms every proposal that needs no typed detail, one after the other; a proposal that needs a phone number, an email address or an amount typed waits on its own card below. Not now dismisses the rest. A morning when nothing happened is skipped for free: no brief, no credits.

lightbulbThree mornings. A dental clinic's Follow-ups agent proposes a thank-you to yesterday's three completed visits and a reschedule text to the one patient who did not arrive. A design agency's Morning brief lists two estimates still open after a week and offers the nudge. A plumber's Missed-call text-back agent already texted last night's two missed callers through its linked automation, so the brief only proposes the callback task.

Probation, then a standing approval

A new agent asks before every send. After you have approved 10 of its sends without editing them, you can grant it a standing approval for one of its templates by typing the agent's name: from then on it sends that exact wording on its own, under its limits - the brief agents send up to 10 messages a day, one per person per day, three per person per week, and stop for good once the person replies. A free-written message, money, team, phone numbers and automation changes always ask.

Linked automations and the no-loop promise

The Missed-call text-back and Review request agents create the automation they need on their first run - the card says so before you turn them on: "Creates: Missed call text-back on first run". It is an ordinary automation under Automations, named after the agent; it pauses when the agent pauses, resumes when it resumes and goes when the agent is deleted, and the agent's settings list it under Linked automations. An agent may also propose creating, changing, pausing or running an automation as a card you confirm.

infoThe no-loop promise: a text an agent sends never starts an automation, and an automation an agent created never wakes that agent again. Both honor the same opt-out list and the same messaging hours, and an agent never proposes the same person twice in one brief or a person who already heard from an agent today.

Two agents that keep the week on track

Two more ready-made agents work the hours before your appointments and the minutes after a message lands. Both only propose: every card waits for a person, and neither one sends a message of its own.

Appointment coordinator
Every morning at your preparation time (08:00 in your organization's time zone by default; change it on the agent's Preparation check row) it reads the next 24 hours: each appointment, its client, who is assigned and what the client wrote lately. It also wakes when an appointment is moved or canceled. When a client names a new date and time it proposes the move as a card; an unclear request ("can we move it?") becomes a task for the appointment owner; a missing phone or email becomes a task with the record link; "see you then" becomes a note. The 4-hour reminder text is the Appointment reminder automation the agent links on its first run - the wording, the date and the time live on that rule, and the agent never writes a reminder of its own. A moved appointment drops its old reminder and gets a new one; a canceled one drops it. It never moves, books or cancels an appointment, never promises availability, and never names a visit type or a health detail. Its number: appointments checked this week, with the scheduling requests still waiting.
Inbox triage
When a client's message lands, and every 30 minutes between 08:00 and 19:00 in your organization's time zone over the open threads still waiting on you, it reads the conversation, finds the client and the thread owner, sorts the request (scheduling, billing, service, complaint or general), judges how urgent it is and proposes one task per conversation: a two-line summary, a priority, a due time, the link and the owner, with a suggested reply written into the task as a draft for a person to send. A complaint, a disputed bill or an unclear request goes to a person with a short acknowledgment only. It never replies, marks a thread read, archives or reassigns it - those are not on its list - and it stops the moment a team member answers. Its number: conversations flagged this week, with how many still need a reply. Up to 48 runs a day, never two within 15 minutes; a quiet half hour is skipped for free.
lightbulbA dental clinic's Appointment coordinator proposes moving tomorrow's 2 PM visit to Thursday because the patient texted the new time, and a task to call the one patient with no phone on file. A property manager's Inbox triage turns "the heater is still not working" into an urgent service task for the on-call technician with a two-line acknowledgment ready to send.

Agents on a project

Any agent can work for one Genaya AI project. Open the project, press the Agents card and Add agent - one of your agents that is not on a project yet, or a ready-made one turned on for the project - or say it in the project's chat: "add the Follow-ups agent to this project". A project agent reads the project's brief, records and rules on every run, the Morning brief and Follow-ups agents read only that project's yesterday and today, its runs show under the project's Activity and its cards wait in the project's decisions. A project holds up to 5 agents and an agent works for one project at a time. See "Add agents to a project" under Projects.

What it costs

A completed agent run is 20 credits from the same Genaya AI pool as a conversation - a credit is one cent of Genaya AI work. A run that looks and finds nothing to do is free, and so is a failed run. Texts and emails an agent sends are counted like any send.

Frequently asked

Was this article helpful?