Β 
In this tutorial, we describe how you can create and configure a new AI Agent, including the voice it answers phone calls with if your organization uses the phone channel. We also show how the "Instruction load" meter and the "Optimize instructions" label tell you whether an AI Agent has too many instructions, and which steps bring them down. Then you will learn what the "Previous AI model" label means. Finally, you will learn how operators rate an AI Agent's answers and actions in a conversation in the Message Center.
Image without caption
To get started, click on the "AI" icon in the left menu.
Then switch to Settings > AI Agent

Creating a New Agent

Use the Add button to create a new AI Agent.
Image without caption
Give the agent an internal name and make the necessary settings:
  • Style
  • Identity
  • Escalation on poor rating
  • Knowledge Base

Connecting a Knowledge Base

In the last step of the creation form you decide whether the AI Agent answers from your Knowledge Base. A Knowledge Base action is a saved setting that tells the AI which part of your Knowledge Base to search: everything, or only selected folders. The step is called "Knowledge Base" in the step bar and headed "Connect Knowledge Base" on the page: "Choose what your AI searches to answer questions: your entire Knowledge Base or only selected folders."
A new AI Agent starts with "Don’t connect a Knowledge Base" selected, so it does not search your Knowledge Base. Choose a Knowledge Base action if it should.
Under "Knowledge Base action" you see one option for each Knowledge Base action you have created, and "Don’t connect a Knowledge Base" at the end. Under each action's name, a line shows what it searches: "Entire Knowledge Base", a number of folders such as "2 folders", or "Individual folders" if the folder names cannot be shown. For an action with selected folders, click the arrow button next to it to show or hide them. Each folder is listed as a path, for example "Products / Shoes". Click an option to select it.
Image without caption
If the action you need does not exist yet, click "New Knowledge Base action" below the options. The dialog "New Knowledge Base action" opens. Enter an action name, then fill in "When to trigger this action" and "How to execute this action". They tell the AI when to search your Knowledge Base and what to look for.
Then decide what the action searches. The switch "Entire Knowledge Base" lets it search everything. If you switch it off, the list below "Individual folders" shows your folders with checkboxes, and you have to tick at least one.
Image without caption
Click "Save" in the dialog. The dialog closes and the new action is selected for the AI Agent, replacing any Knowledge Base action you had selected. The action stays in your list even if you do not finish the form. "Cancel" closes the dialog without creating anything.
If your role has no access to the Knowledge Base, the switch "Entire Knowledge Base" starts switched on. If you switch it off, the folder list is replaced by the note "The folders can’t be shown because you don’t have access to the Knowledge Base."
The AI Agent itself is only created when you click the "Save" button of the form, below the options.

Settings

After creating the agent, you will return to the agent overview.
Now click on the name of the AI Agent you just created.
You will access the settings of this specific AI Agent:
Image without caption
Here you can now also enrich the welcome message with images, videos, or buttons:
Image without caption
Then switch to the

"Profile" Tab

Here you can now make detailed settings for your agent.
Define
  • what the agent should be called
  • which operator group and time zone it should belong to
  • which conversation timeout should apply to the agent
  • which skills it should master (only in the Flex package)
  • in which behavior mode it should operate
  • which action it should perform when the timeout is reached
  • whether the bot should use the widget's login mode
Further information can be obtained by clicking on the respective information buttons behind the options.

"Voice" Tab

You only see this tab if the phone channel is enabled for your organization. The phone channel is in Alpha state; reach out to your account manager if you want to test it.
Voice agents are the AI Agents that answer your phone calls. Before they can take calls, your organization has to confirm the "Data processing notice for voice agents". Until then, this tab shows the heading "Voice agents not activated" instead of the voices, and your phone channels take no calls; callers hear the channel's message for when the AI Agent is unavailable. The confirmation applies to your whole organization; if your account has several organizations, each one confirms separately. If the notice changes, it has to be confirmed again.
To confirm, your role needs "Edit" for "Chat settings". It also needs "Edit" for "Organizations", unless it is the Owner role. By default, the Owner and Manager roles meet both. If you cannot confirm it yourself, ask an operator whose role can, by default an Owner or a Manager.
If you can confirm, the data processing notice opens by itself when you open the tab. Read it and click "Confirm and activate", and the tab then shows the voices. "Cancel" closes the notice without confirming it.
Image without caption
If you closed the notice with "Cancel", the "Review notice" button opens it again.
Image without caption
If you cannot confirm, you see the heading and one sentence about it, but no button, and cannot open the notice. More on the consent is on the "Phone (Alpha)" page.
Once the notice is confirmed, you choose here the voice your AI Agent answers phone calls with. Every AI Agent starts with a voice already chosen, so you do not have to pick one here for calls to work. The voice belongs to the AI Agent rather than to the phone channel, so it applies everywhere this agent takes calls.
Twelve voices are available. Each card gives the name of the voice, whether it sounds female or male, and two words describing its character, such as "Warm" and "Calm" for Mara or "Confident" and "Steady" for Jonas. The name is only there for picking; your callers never hear it.
Which language a call uses is set on the phone channel, under "Primary call language", not here. The voices on offer speak every call language. If they were made for only some languages, a line above the cards says so, for example "These voices were made for German; other languages may sound less natural." Listen to a preview in your call language to judge.
Listen to a voice before you decide. Under "Listen to preview in" at the top, pick the language you want to hear it in, ideally the one your phone channel takes calls in. Then click the play button on the card, and click the same button again to stop. That choice only applies to the preview and does not change the call language.
To pick a voice, click the card itself rather than its play button. The selected card is highlighted. The "Default" label marks Ida, the voice every AI Agent starts with; it does not show your selection. Pick another voice and the highlight moves while the label stays on Ida. The highlighted card is the voice your agent uses once you save.
Image without caption
Finally, click "Save" at the bottom of the tab. The change applies to calls this agent takes afterwards.
For a call to reach this agent at all, a phone channel has to point at it. That is not set here but on the phone channel itself, under Settings Β» Channels Β» Phone: open the channel, select the agent in its "AI Agent" field and save there.
The button at the top right of this tab opens that page in a new browser tab:
  • If this agent answers exactly one phone channel and your role has "Edit" for "Chat settings", it reads "Phone channel settings" and opens that channel.
  • In every other case, also for a new agent with no channel yet, it reads "Phone channels" and opens the list.
  • Without at least "View" for "Chat settings", the button does not show.
The tab shows the message "The available voices could not be loaded. Please try again." The list of voices could not be loaded. Reload the page. If the message stays, contact our support team.
The tab shows the message "Voice previews are not set up for this channel's voice engine." You can still pick and save a voice. Only listening to a sample is unavailable for this agent. The voice engine is set up by us, not in your settings, so you cannot change this yourself; contact our support team if you need previews.

"Data Access" Tab

Select which of Lime Connect's data fields should be available to the AI when answering customer inquiries.

"Guidelines" Tab

Define which general behavioral instructions your AI Agent should follow.
Click a guideline's name to edit it. When you save or cancel, you return to this AI Agent.

"Actions" Tab

Here you can select which of the created system actions your AI Agent should master.

How do I know if an AI Agent has too many instructions?

An AI Agent has too many instructions when its "Instruction load" meter is above 100%. The meter is on every tab of the AI Agent's settings, under its name. If the AI Agent still uses our previous AI model, a notice about its switch to the new model appears above the meter (see "What does "Previous AI model" mean?" below).
The meter measures the length in characters of all the instructions the AI Agent has to keep in mind for every answer and shows it as a percentage of the recommended amount. 100% is that recommended amount: it is fixed, the same for every AI Agent, and not a limit you set. The more instructions there are, the less reliably the AI follows each rule.
The legend under the bar shows where the instructions come from. Each part is shown as a percentage of the recommended amount, so the three parts add up to the total above the bar:
  • Guidelines: the text of the guidelines switched on in the AI Agent's "Guidelines" tab. Every organization gets a short pre-configured guideline, and it is switched on for every new AI Agent, so a new AI Agent never starts at 0%.
  • Actions: the "When to trigger this action" and "How to execute this action" fields of the actions switched on in the AI Agent's "Actions" tab. That tab switches actions on and off for this AI Agent; the "Actions" menu under Connect AI (the "AI" icon in the left menu) lists all your actions by type, and that is where you edit an action. Actions marked "Blocked" do not count: Lime Connect has blocked them, the AI does not use them, and you cannot block or unblock an action yourself. The "Blocked" tag appears next to the action's name, in the "Actions" tab and in the "Actions" menu.
  • Brand and tone: fields in the "General" tab, the first tab, which opens when you click the AI Agent's name:
  • "Brand description" in the "Identity" section. It always counts.
  • "Custom role instructions" in the "Style" section. They count only when "Role" is set to "Custom".
  • "Custom tonality instructions" in the "Style" section. They count only when "Tonality and communication style" is set to "Custom".
Nothing else counts, neither the "Data access" tab nor the knowledge base.
At or below 100%, the meter stays blue and gray. Above 100%, the percentage turns orange, a marker shows where 100% is, and an orange bar marks the part beyond it.
The meter never blocks anything (unlike the limit of 20 guidelines per AI Agent, described in the help article "Guidelines"): you can always save the AI Agent.
Above 100%, the warning note "The AI follows each rule less reliably when it has a lot to keep in mind. Trim or merge guidelines to bring this down." also appears under the meter. Clicking "guidelines" in that warning note opens this AI Agent's "Guidelines" tab. The warning note always names guidelines, but start with the largest part of the legend:
  • Guidelines: in the "Guidelines" menu under Connect AI, where guidelines are created and edited, shorten long guidelines or merge similar ones. A change applies everywhere the guideline is switched on. Pre-configured guidelines cannot be edited. In the AI Agent's "Guidelines" tab, switch off the ones it does not need.
  • Actions: in the "Actions" menu under Connect AI, shorten the "When to trigger this action" and "How to execute this action" fields. A change applies everywhere the action is switched on. If only this AI Agent does not need an action, switch it off in its "Actions" tab instead.
  • Brand and tone: in the "General" tab, shorten the "Brand description" or your custom role and tonality instructions.
Work through the list until the percentage is 100% or below. As soon as you save a change, the meter shows the new value.
The help article "Guidelines" (not the "Guidelines" tab) gives the details in the section "How do I know if a setup has too many instructions?". That article, like the "Setups" menu under Connect AI, calls an AI Agent or a Copilot a "setup".
You can also spot an AI Agent above 100% in the overview under Connect AI > Setups > AI Agent: an orange "Optimize instructions" label appears next to its name. Hover over the label to see the percentage. The label itself is not a link: click the AI Agent's name to open its settings and see the meter. The blue "Previous AI model" label, which marks an AI Agent that still uses our previous AI model, takes the place of "Optimize instructions", so if you see it, open the AI Agent and check its meter.

What does "Previous AI model" mean?

In the overview under Setups > AI Agent, the blue "Previous AI model" label can appear next to an AI Agent's name. That AI Agent still works with our previous AI model. New AI Agents use the new, more powerful model automatically and do not carry the label. If none of your AI Agents shows the label, this section does not concern you.
Hover over the label to see a notice with the date on which we switch this AI Agent to the new model.
Image without caption
The same notice appears at the top of the settings of an AI Agent with the label, on every tab. You do not have to do anything: we switch the AI Agent on that date, your settings stay as they are, and the label and the notice disappear after the switch. There is no setting to switch the model yourself.
If you want to see how your AI Agent answers with the new model before the switch date, create a second AI Agent with the same settings, as described above under "Creating a New Agent" and "Settings", and try it out.

How do I rate AI answers and actions in the Message Center?

You can rate what an AI Agent did in a conversation: each answer it wrote and each action it triggered. A rating says whether the answer or action was right, and you can add a reason and a comment.
To add a rating, your role needs "Edit" for "Administration" in the Connect AI permissions, and your organization's plan must include the AI Agent product. With less, you still see the ratings your colleagues left, but you get no thumbs to add one.
Open a conversation in the Message Center in which an AI Agent answered or triggered an action. Below each answer of the AI Agent, and below each action it triggered (the row "AI action triggered" with the card of the action), you see a thumbs-up button and a thumbs-down button. These thumbs are the operator rating described here. If a contact rated an AI answer too, that rating is shown separately, and it is not the same as the "Escalation on poor rating" setting, which reacts to a contact's rating.
Image without caption
Click the thumbs-up button if the answer or action was right, and the thumbs-down button if it was not. The rating is saved right away, and the "Thank you!" dialog opens with the line "Your feedback helps improve AI responses". Whatever else you add in the dialog is kept only after you click "Save changes". If saving fails, an error message appears. The dialog stays open, so you can try again.
In the dialog you can add detail, all of it optional:
  • "Positive" and "Negative": switch the rating if you clicked the wrong thumb.
  • "Select reason": appears only for "Negative". Choose the closest one.
  • For an AI answer: "Incorrect facts", "Missing information", "Incorrect / missing links", "Wrong language", "Off topic / irrelevant", "Poor style / phrasing", "Wrong / no action triggered" or "Harmful content".
  • For an AI action: "Wrong moment", "Did not execute", "Wrong result" or "Not needed".
  • "Describe your experience": a free text field with the hint "Share details (optional)".
Image without caption
"Save changes" appears as soon as you change something in the dialog. "Close" keeps the rating you clicked but drops anything you did not save.
Once an answer or action is rated, the thumbs are replaced by a small badge below it. It shows a filled thumb, and next to it the reason, or the comment if you gave no reason. It also shows the picture of the operator who rated; hover over the picture to see the name and the date. Every operator who opens the conversation sees the same badge.
Image without caption
To change a rating later, click the badge. The dialog opens again with the operator's name ("Created by") and the date at the bottom; change what you need and click "Save changes".
You rate actions only in the conversation view of the Message Center, not in the execution history of an AI Agent.