AI agents

In short: An AI agent is who does the work: a name, a description, instructions, knowledge, memory and skills. Agents belong to the whole account. What kind of work an agent does (write a reply, choose from a list, carry out a task) is decided by aishaa per workflow step; which agent answers customer messages is decided by the organisation’s workflow with the Write the reply step. Every reply from an agent starts as a proposal to a colleague.

Where to find this

Go to Account > Agents. The list of agents is on the left, the opened agent with its tabs on the right. The explanation at the top reads Configure AI agents and control their behavior, tools, and permissions.

Who: changing agents is possible from the Account admin role. An Organisation admin can view the page; a notice at the top then says that agents are shared with the whole account and that only account admins can change them, and the buttons to create and save are missing. Agents apply to the whole account.

The Agents page with, on the left, the list of agents and, on the right, an opened agent with the Answers for chips and the plugins, and the tabs Basics & behavior, Skills, Knowledge, Trust and Version history, menu Account > Agents.

Agent, skills and tasks

aishaa makes a clear distinction:

So an agent is not a fixed role. The same agent can write replies in one workflow and choose or carry out a task in another workflow.

The list of agents

• Find an agent with Search agents. You search by name and description.
• A disabled agent has the chip Disabled.
• If there are no agents yet, you see No agents yet with Create your first AI agent to start configuring intelligent behavior.
• Click New agent to add an agent, or Library to choose a ready-made agent. See The library.

Above the tabs of an opened agent there are chips:

Creating an agent

1. Click New agent.
2. Under Start from, choose An empty agent or an agent from the library. An agent from the library fills in the name, description, instructions, memory and skills for you.
3. Fill in the fields (see below) and click Save. You see Agent created successfully.

A new agent starts with Enabled switched on and with the skills of all connected plugins. You can only save when Agent name and General instructions are filled in. If you leave an agent with unsaved changes, aishaa asks Discard the unsaved changes to this agent?

Basics & behavior tab

The Basics & behavior tab (Define the core identity and behavior of this agent.) contains:

If the agent was made from an agent in the library, it says Made from: (name) with the Back to the standard button. That button sets the description and instructions back to the text from the library; the name, tools and rules stay. See The library.

Putting the organisation name in the instructions

One agent can work for several organisations. If you want the agent to mention the name of the right organisation, use the variable instead of a fixed name.

Below General instructions is the Insert variable: bar with the Organisation name chip. One click puts the variable where your cursor is. The tooltip explains: aishaa fills in the brand name of the organisation the conversation belongs to. aishaa uses the organisation’s Brand name, or the organisation name if no brand name has been filled in. See Organisation information.

If you type something between double curly brackets yourself that aishaa does not know, a warning appears that aishaa does not know that variable and that the text stays exactly as written.

Skills tab

On the Skills tab you choose what the agent can do: which skills it has, and per tool whether it may use it, whether it may only propose an action or carry it out itself, and under which conditions (channel, new or existing customer, attachment, customer group, within or outside opening hours). An agent only gets tools through its skills. This tab is described separately in Skills.

Knowledge tab

The Knowledge tab has two blocks.

The Knowledge tab of an agent with the choice between All knowledge and Chosen knowledge, the tags and documents, and below it the Memory block with three switches, menu Account > Agents.

Knowledge: what the agent searches

What this agent looks up when it answers. It searches itself, as often as it needs, and only in the knowledge chosen here.

With Chosen knowledge:

• Under Tags you choose tags. A tag takes every document that carries it, also documents you add later.
• Under Documents (0 chosen) you tick individual documents. Use Search documents to search by name. Each document shows which organisation it applies to (or Whole account) and can be sent as attachment if the agent can send it along as an attachment.
• If you choose nothing, you see Nothing chosen yet: this agent will not search any knowledge.

See Knowledge documents.

Memory: what the agent remembers

What this agent remembers beyond the conversation. The organisation’s rules per topic and colleagues’ notes about a customer always reach the agent.

All three are on by default. Turn off what an agent may not use, for example the customer’s earlier conversations for an agent that only answers general questions.

Trust tab

The Trust tab appears for a saved agent. It shows, per organisation and topic, how often colleagues sent this agent’s proposals unchanged. Where the agent has built up enough trust, an Account admin can let it send by itself on that topic with Let it send by itself; with the Emergency stop you set everything back to proposals per organisation. See Trust and letting an agent send by itself.

Version history tab

Every time an agent is changed or restored, aishaa first saves the previous version automatically. The Version history tab (Earlier versions that were automatically saved before this agent was changed.) shows for each version the date and time, the reason (before update or before restore) and the colleague who made the change.

1. View snapshot opens a window with the settings of that version.
2. Restore snapshot restores that version. Confirm the question Are you sure you want to restore ‘(name)’ from (date)?; you then see Snapshot restored. Restoring is not possible while there are unsaved changes, while you are creating a new agent, or for a locked agent.

If nothing has been saved yet, it says No snapshots for this agent yet.

Duplicating, deleting and locking

At the bottom of the agent are Duplicate, Delete, Cancel and Save. Duplicating and deleting only work for a saved agent without unsaved changes.

Deleting does not work in these cases:

• A workflow step names the agent: Een workflowstap noemt deze agent. Kies daar eerst een andere agent.
• It is the only enabled agent of the account: Dit is de enige actieve agent van het account en kan daarom niet worden verwijderd.
• The agent is locked.

Above the tabs is the Lock button. Nobody can change, delete or restore a locked agent any more, not even aishaa and the assistant, until an Account admin unlocks it. See Lock configuration.

Which agent answers

You do not choose which agent answers a customer message on this page but in the organisation’s workflows. Every organisation has a workflow that carries out a Write the reply step on every understood customer message; the agent that step names writes the reply. An organisation automatically gets the standard workflow Answering customer messages with its first message, if no workflow of the organisation answers yet. See Answering customer messages with workflows.

If your account has several organisations and you are an Account admin, the Who answers where section is at the bottom of the page (The workflows of each organisation decide which agent answers its customers: the agent their “Write the reply” step names. You change it in that workflow.).

The Who answers where section at the bottom of the Agents page, with for each organisation the agent that answers, the Change in workflow button and the plugins, menu Account > Agents.

If several workflows answer the same message, each workflow writes its own proposal with the agent its step names. In this overview, and for work where no agent is named, the agent of the oldest active answering workflow that names an enabled agent applies.

Who continues working on a conversation

Work on a conversation continues with the agent that wrote the last reply, not with “the organisation’s agent”. If there is no reply yet, the agent named by the answering workflow works on it. This applies to everything a colleague asks aishaa in the ticket:

• Run analysis again in the top bar: aishaa looks at her work on the latest message again and picks up where she was. Under Extra instructions (optional) you tell her what to do differently. The message is not read again; the agent continues with what it already knew and writes a new version of its proposal.
• Handle anyway on a message that aishaa thought needed no reply: the agent then writes a proposal after all. See No reply needed.
• Adjusting the proposal with instructions in the editor, and asking questions in aishaa on this ticket. See aishaa on this ticket.

If the last message is yours, you see The latest message is ours, so there is nothing for aishaa to answer. If the agent is still working on the message, you see aishaa is still working on this message. Ask again when she is done.

Good to know

• Every reply from an agent is a proposal that a colleague checks, until someone lets the agent send by itself on a topic. See Trust and letting an agent send by itself.
• You do not set the writing style (tone, structure) per agent but per channel, see Writing styles. Put the rules about content in the General instructions.
• Rules that apply to one organisation only go under Additional instructions on the Organisation information page of that organisation, see Organisation information. If you really want different behaviour per organisation, create a second agent and name it in that organisation’s answering workflow.
• The agent itself searches the knowledge, earlier answers and catalogues from connected systems, within what you allow on the Knowledge tab and under Memory.
• You can also have the General instructions changed by asking aishaa, see Tell aishaa. Such a change always comes back as a proposal.
• All changes to agents appear in the Changes overview under the type AI agent. See Changes and reverting.
• A change to an agent’s instructions, knowledge, memory or skills makes the built-up trust start again. A new name or description does not.

Related

• Skills
• Trust and letting an agent send by itself
• The library
• How aishaa processes a ticket
• Answering customer messages with workflows
• Organisation information
• Writing styles
• Knowledge documents
• What aishaa knows
• Tell aishaa
• Changes and reverting
• Lock configuration