Skip to content

Flows

Flows that know when to ask.

A flow is a fixed series of steps. AI agents do the work, and a person decides at the checkpoint.

  1. Trigger
    New ticket in Zendesk
    When a customer writes
  2. AI agent step
    Iris looks up the order
    Iris · Support desk
  3. Condition
    Is it about money?
    yes
    Iris decidesno
  4. Checkpoint
    Approve the refund
    approved
    Reviewer: Jorienrejected
  5. Integration action
    Refund in Mollie
    Create a refund as Iris
  6. Answer
    Iris answers the question
    In the Zendesk conversation

Ready to run. Press Run now to start one.

Model choicePer step, within your policyRouting
Cost
–
Latency
–
In
–
Out
–
Sample data

Step 1 of 7: The flow, built

In one sentence

A flow is a multi-step process in which AI agents do the steps and a person decides at the checkpoints. It runs the same way every time, which a single instruction does not.

A published graph of steps: AI agent steps, conditions, fan-outs, checkpoints where a person decides and integration actions with fixed arguments. Runs stay on the version they started on and carry their own budget.

How it works

Every step in plain words.

The graph is fixed. AI agents work inside it and never redraw it.

  • Steps you can read

    AI agent steps, conditions, a step for each item in a list, integration actions with fixed arguments, and checkpoints. The graph is the documentation.

  • A person at the checkpoint

    A checkpoint pauses the run and asks a person to decide before it continues. The decision is recorded with the run.

  • Started by what happens

    A schedule, a webhook, a watch on a system you run, a finished task, a change in the knowledge base or a customer conversation on one of your channels can start a flow.

  • Versioned

    Publish a new version and new runs follow it. Runs already underway finish on the version they started on.

  • Its own budget

    A run carries its own budget, so one busy flow cannot spend the month on its own.

Triggers

Six ways to start.

Everything an AI agent does starts as a task or as a flow run. The only question is what started it.

  • A schedule

    A cron expression, checked every minute. A slow run is never overtaken by the next one.

  • A webhook

    An address nobody can guess, from anything that speaks HTTP.

  • A watch on one of your systems

    Reads an integration on a schedule and starts the flow only when the result has changed.

  • A finished task

    One task finishes and the next flow starts. Tasks come from the app, the API, chat or Pal.

  • A change in your knowledge

    When a knowledge source changes, that change can start a flow.

  • A customer on a channel

    Your website, WhatsApp, email, Genesys or your own app through the channel API.

Channels

A customer message starts a flow.

Your channels start flows. When a customer needs a person, the AI agent hands the conversation to your contact centre (Genesys and others), with its context.

  • Hands over to
  • Genesys Cloud
  • Talkdesk
  • Intercom
Channels · website chatExample

You are talking to an AI agent

Where is my order #20417?

It left the warehouse this morning. The track-and-trace link is in your email.

Answered by AI · source: Shipping

It was meant as a gift. Can I talk to someone?

Handed to your contact centre, with the transcript

  1. Trigger
    A customer on your website (Completed)
    Website chat
  2. Answer
    Nova answers from your knowledge (Completed)
    Knowledge: Shipping
  3. Condition
    Needs a person? (Completed)
    yes
    Nova decidesno
  4. Hand over
    To your contact centre (Completed)
    Genesys Cloud · with the transcript
  • Every channel starts a flow

    Your website, WhatsApp, email, Genesys Messenger or your own app through the channel API. A customer conversation is a trigger like any other.

  • The flow picks who answers

    An answer step names the AI agent that replies, from the knowledge you choose, and a condition sends each question to the right one.

  • A hand-over with its context

    When a customer needs a person, a hand-over step passes the conversation to your contact centre (Genesys Cloud, Talkdesk or Intercom) with the transcript so far.

  • Honest about being an AI

    An AI agent on a channel never claims to be human, your website and the channel API show an AI notice, and channels run only on the models you allowed for public use.

  • The same controls

    Anything a channel AI agent would change in a connected system follows the same approval list as everywhere else, and every conversation is on the record.

Where it stops

  • Flows. A flow runs the graph you published: AI agents work inside it and never redraw it. A single standing instruction with a trigger is an automation, not a flow.
  • Channels. A channel flow decides who answers. What an AI agent may change on a customer's behalf is decided by the approval list, not by how the question was phrased.
How is a flow different from an automation?

An automation is one standing instruction with a trigger. A flow is several steps with conditions and checkpoints, in a fixed order.

Can a flow act in our systems without an AI agent?

Yes: an integration action runs a connector action with fixed arguments. If that action is on the approval list, it asks a person in the flow as it would in chat.

Which channels can start a flow?

Your website, WhatsApp, email and Genesys Messenger, and your own app through the channel API. Each conversation starts the flow its channel names.

Where does a customer go when they need a person?

To your own contact centre. The hand-over step passes the conversation to Genesys Cloud, Talkdesk or Intercom with the transcript, so your people pick it up where the AI agent left off.

See it at work in your organisation.

We start by listening to your question. Then we show you a demo that fits your organisation.