KIBO AI is a conversational assistant built directly into the KIBO Admin UI, giving us a way to complete platform tasks by describing what we want in plain language instead of navigating through menus and forms manually. It’s “bring-your-own-model”, meaning we connect the LLM provider and credentials of our choice, and KIBO AI uses that connection to understand requests, reason through them, and carry out actions across the platform on our behalf.In this video, we’ll set up a model, configure how it’s used across the platform, and then walk through a real example, creating a promotion, to show how KIBO AI asks clarifying questions, explains its own reasoning, confirms with us before taking any action, and only makes changes we’ve approved.Important note: KIBO AI requires at least one connected and validated model credential before it can be used; until that step is complete, we’ll see a “Connect a model to get started” prompt instead of an active assistant. KIBO AI is a conversational assistant built directly into the KIBO Admin UI, giving us a way to complete platform tasks by describing what we want in plain language instead of navigating through menus and forms manually. It’s “bring-your-own-model”, meaning we connect the LLM provider and credentials of our choice, and KIBO AI uses that connection to understand requests, reason through them, and carry out actions across the platform on our behalf. In this video, we’ll set up a model, configure how it’s used across the platform, and then walk through a real example, creating a promotion, to show how KIBO AI asks clarifying questions, explains its own reasoning, confirms with us before taking any action, and only makes changes we’ve approved. Important note: KIBO AI requires at least one connected and validated model credential before it can be used; until that step is complete, we’ll see a “Connect a model to get started” prompt instead of an active assistant. This is the KIBO Admin UI. To invoke the KIBO AI interface, we’ll click the round KIBO logo on the bottom right of the screen. This is the KIBO Admin UI. To invoke the KIBO AI interface, we’ll click the round KIBO logo on the bottom right of the screen. If this is the first time we’re using KIBO AI, we’ll need to configure the LLM model we’re going to use with KIBO. This is a simple and straightforward process we’ll walkthrough right now. In the KIBO AI side panel, we can click the “Configure a model” button to be taken directly to the “Agent Models” configuration menu. We can also click the “Read the setup guide” link, which will open the KIBO AI documentation for us to review in a separate tab. If this is the first time we’re using KIBO AI, we’ll need to configure the LLM model we’re going to use with KIBO. This is a simple and straightforward process we’ll walkthrough right now. In the KIBO AI side panel, we can click the “Configure a model” button to be taken directly to the “Agent Models” configuration menu. We can also click the “Read the setup guide” link, which will open the KIBO AI documentation for us to review in a separate tab. Alternatively, we could access the “Agent Models” configuration menu by navigating to the “SYSTEM” tab in the left-hand menu, clicking on “Settings”, and then clicking “Agent Models”. Alternatively, we could access the “Agent Models” configuration menu by navigating to the “SYSTEM” tab in the left-hand menu, clicking on “Settings”, and then clicking “Agent Models”. The “Agent Models” configuration menu opens on the “Slot Configuration” screen. This is where we determine what models we’re using for either “Chat” or “Image” generation. We’ll discuss this screen further shortly. The “Agent Models” configuration menu opens on the “Slot Configuration” screen. This is where we determine what models we’re using for either “Chat” or “Image” generation. We’ll discuss this screen further shortly. For now, we’ll click the “Credentials” button on the top right. For now, we’ll click the “Credentials” button on the top right. The “Credentials” screen displays every model connection that has been previously added, along with its “Provider”, “Default Model”, supported “Modalities”, and current “Status”. This is also where we go to add a new credential, edit one, or remove one we no longer need. The “Credentials” screen displays every model connection that has been previously added, along with its “Provider”, “Default Model”, supported “Modalities”, and current “Status”. This is also where we go to add a new credential, edit one, or remove one we no longer need. To add a new LLM Model to our KIBO tenant, we’ll click the “Add Credential” button on the top right. To add a new LLM Model to our KIBO tenant, we’ll click the “Add Credential” button on the top right. Clicking “Add Credential” opens a blank form where we’ll enter the information for the LLM Model we’ll be using. All required fields are noted with a “red star”. Clicking “Add Credential” opens a blank form where we’ll enter the information for the LLM Model we’ll be using. All required fields are noted with a “red star”. We’ll start by giving the credential a “Name” so we can identify it later, here entered as “New Credential”. We’ll start by giving the credential a “Name” so we can identify it later, here entered as “New Credential”. Since KIBO AI is defined as “bring-your-own-model”, the “Provider” dropdown isn’t limited to one company: we can choose “openai”, “anthropic”, “google”, or “other” for any provider not listed here. Since KIBO AI is defined as “bring-your-own-model”, the “Provider” dropdown isn’t limited to one company: we can choose “openai”, “anthropic”, “google”, or “other” for any provider not listed here. In this example, we’ll select “anthropic” as the provider. The “Default Model” field will then display the models available for Anthropic, and we’ll pick whichever best fits the task and budget for this credential. In this example, we’ll select “anthropic” as the provider. The “Default Model” field will then display the models available for Anthropic, and we’ll pick whichever best fits the task and budget for this credential. If we select “other” instead of a named provider, we’ll hand enter the “Default Model” we’re using. KIBO AI will now require a “Base URL” to know where to send requests. This is important: KIBO AI won’t have anywhere to route the request until that URL is filled in, and the form flags it until we do. If we select “other” instead of a named provider, we’ll hand enter the “Default Model” we’re using. KIBO AI will now require a “Base URL” to know where to send requests. This is important: KIBO AI won’t have anywhere to route the request until that URL is filled in, and the form flags it until we do. We’ll complete the LLM Model configuration by adding in the “API Key” for the provider account. We’ll also select whether we are using this model for “Chat”, “Image”, or both.Important note: the “API Key” is stored securely and is never shown again once saved, so it’s worth double-checking it before moving on. We’ll complete the LLM Model configuration by adding in the “API Key” for the provider account. We’ll also select whether we are using this model for “Chat”, “Image”, or both. Important note: the “API Key” is stored securely and is never shown again once saved, so it’s worth double-checking it before moving on. Before saving, we can click “Test” to confirm the credential actually works. Catching a bad key here, before it’s saved and assigned to a slot, keeps a broken credential from quietly failing later. Before saving, we can click “Test” to confirm the credential actually works. Catching a bad key here, before it’s saved and assigned to a slot, keeps a broken credential from quietly failing later. Once we’ve added in all of the LLM Model configuration information, and performed a valid “Test”, we’ll click the “Save” button on the top right. Once we’ve added in all of the LLM Model configuration information, and performed a valid “Test”, we’ll click the “Save” button on the top right. We can also test the connection of any existing models in the credential list. Clicking “Test” will run the connection test and return the same information we just saw on the previous step. We can also test the connection of any existing models in the credential list. Clicking “Test” will run the connection test and return the same information we just saw on the previous step. Clicking the “ellipses” on the far right of an existing connection allows us to “Edit” or “Delete” the connection. Clicking the “ellipses” on the far right of an existing connection allows us to “Edit” or “Delete” the connection. Clicking “Edit”, or directly on any of the existing connections allows us to edit it as needed. Again, we’ll click “Save” on the top right when we’ve finished editing. Clicking “Edit”, or directly on any of the existing connections allows us to edit it as needed. Again, we’ll click “Save” on the top right when we’ve finished editing. Back on the “Slot Configuration” page, we can set the model for either “Thinking” or “Flash”, which are “Chat” options, or for “Image”, using the “Active Credential” dropdown. We can also define the “Model Override” and “Monthly Token Limit”, if they are applicable, and can again test the connection of the Model. When we’ve finished configuring the slots, we’ll click “Save Configuration” on the left. Back on the “Slot Configuration” page, we can set the model for either “Thinking” or “Flash”, which are “Chat” options, or for “Image”, using the “Active Credential” dropdown. We can also define the “Model Override” and “Monthly Token Limit”, if they are applicable, and can again test the connection of the Model. When we’ve finished configuring the slots, we’ll click “Save Configuration” on the left. The “Token Usage” screen lets us track how much we’re using the selected LLM Models over time. We can set a custom “Start Date” and “End Date”, or use the “Last 7 Days”, “Last 30 Days”, or “Last 90 Days” shortcuts. We can also choose to view Token Usage by “Credential” or “Model”, and by “Day” or “Cumulative”. This example shows two credentials by day over the last 30 days. The “Token Usage” screen lets us track how much we’re using the selected LLM Models over time. We can set a custom “Start Date” and “End Date”, or use the “Last 7 Days”, “Last 30 Days”, or “Last 90 Days” shortcuts. We can also choose to view Token Usage by “Credential” or “Model”, and by “Day” or “Cumulative”. This example shows two credentials by day over the last 30 days. This view displays “Token Usage” by model and cumulative usage over the same 30 day timeframe. This view displays “Token Usage” by model and cumulative usage over the same 30 day timeframe. Back on the KIBO Admin UI homescreen, with a model connected and configured, we can open KIBO AI by clicking the KIBO logo on the bottom right.This is the “Just ask KIBO AI” entry point, where we can either type a request directly or click “Choose an area” to browse by capability. Back on the KIBO Admin UI homescreen, with a model connected and configured, we can open KIBO AI by clicking the KIBO logo on the bottom right. This is the “Just ask KIBO AI” entry point, where we can either type a request directly or click “Choose an area” to browse by capability. Clicking “Choose an area” expands a list of six capability areas KIBO AI can help with: “Analytics”, “Orders & Routing”, “Returns & Reverse Logistics”, “Search & Merchandising”, “Products & Categories”, and “Promotions”, each covering a different part of the platform. Clicking “Choose an area” expands a list of six capability areas KIBO AI can help with: “Analytics”, “Orders & Routing”, “Returns & Reverse Logistics”, “Search & Merchandising”, “Products & Categories”, and “Promotions”, each covering a different part of the platform. Selecting an “area”, in this case “Promotions”, surfaces a handful of example prompts to help us get started. Selecting an “area”, in this case “Promotions”, surfaces a handful of example prompts to help us get started. These examples are just a starting point, though; we can always type our own request in the “Ask KIBO AI anything” box at the bottom of the panel instead, tailored to whatever our business actually needs, regardless of any selected “Capability”. These examples are just a starting point, though; we can always type our own request in the “Ask KIBO AI anything” box at the bottom of the panel instead, tailored to whatever our business actually needs, regardless of any selected “Capability”. In this example, we’ve selected the example Promotions prompt “Create free shipping over a minimum order amount”. The model “Thought for 5 seconds”, then surfaces some questions about the free shipping discount we want to create. In this example, we’ve selected the example Promotions prompt “Create free shipping over a minimum order amount”. The model “Thought for 5 seconds”, then surfaces some questions about the free shipping discount we want to create. Clicking into the “Thought for 5 seconds” label expands KIBO AI’s reasoning behind that response. This is the transparency the thinking model is built to provide, letting us see how the assistant is working through a request rather than treating it as a black box. Clicking into the “Thought for 5 seconds” label expands KIBO AI’s reasoning behind that response. This is the transparency the thinking model is built to provide, letting us see how the assistant is working through a request rather than treating it as a black box. We don’t need a separate form to answer KIBO AI’s questions; we can just type the details, like the discount “Name”, “Minimum Order Amount”, and “Start” and “End Dates”, right back into the same chat box, and KIBO AI picks up the conversation from there. We don’t need a separate form to answer KIBO AI’s questions; we can just type the details, like the discount “Name”, “Minimum Order Amount”, and “Start” and “End Dates”, right back into the same chat box, and KIBO AI picks up the conversation from there. Once it has what it needs, KIBO AI recaps the full discount configuration and asks “Does this look correct?” with “Yes, create it” and “No, let me adjust” options.Important note: KIBO AI will always verify before taking an action like this, so nothing gets created without an explicit approval. Once it has what it needs, KIBO AI recaps the full discount configuration and asks “Does this look correct?” with “Yes, create it” and “No, let me adjust” options. Important note: KIBO AI will always verify before taking an action like this, so nothing gets created without an explicit approval. After we click “Yes, create it”, we can see KIBO AI actually run a tool, shown here as “Navigate To Page”, to open the discount creation screen for us. This visibility into which tools are being used, rather than hiding the mechanics, is part of the same transparency built into the thinking model. After we click “Yes, create it”, we can see KIBO AI actually run a tool, shown here as “Navigate To Page”, to open the discount creation screen for us. This visibility into which tools are being used, rather than hiding the mechanics, is part of the same transparency built into the thinking model. KIBO AI lands us on the “Create Discount” page with all the fields already filled in based on our conversation, and it asks us to review the details and click “Save”.Important note: KIBO AI fills in the recommended configuration for us, but it stops short of clicking “Save” itself, keeping that final action in our hands. KIBO AI lands us on the “Create Discount” page with all the fields already filled in based on our conversation, and it asks us to review the details and click “Save”. Important note: KIBO AI fills in the recommended configuration for us, but it stops short of clicking “Save” itself, keeping that final action in our hands. At the top of the KIBO AI chat panel, we can click the ”+” icon to start a new chat at anytime. This will give us the same “Choose an area” dropdown as well as the “Ask KIBO AI anything” chat box. At the top of the KIBO AI chat panel, we can click the ”+” icon to start a new chat at anytime. This will give us the same “Choose an area” dropdown as well as the “Ask KIBO AI anything” chat box. Clicking the “Chat Bubbles” icon shows every previous chat we’ve had with KIBO AI, complete with a preview of each one and a search bar to find a specific conversation. We can also click the “New chat” button if we want to start something new. Clicking the “Chat Bubbles” icon shows every previous chat we’ve had with KIBO AI, complete with a preview of each one and a search bar to find a specific conversation. We can also click the “New chat” button if we want to start something new. Typing keywords into the search bar quickly locates specific chats. Typing keywords into the search bar quickly locates specific chats. And for each conversation, clicking the “Ellipses” on the far right allows us to copy the “Correlation ID” of the chat for troubleshooting purposes, “Rename” the chat for easier recognition, and “Delete” the chat entirely if we no longer need it. And for each conversation, clicking the “Ellipses” on the far right allows us to copy the “Correlation ID” of the chat for troubleshooting purposes, “Rename” the chat for easier recognition, and “Delete” the chat entirely if we no longer need it. Clicking back into the conversation at the top of the list picks up exactly where we left off, recap and all. Any past conversation in the list can be reopened this way, not just the most recent one, so nothing gets lost if we step away or switch tasks partway through. Clicking back into the conversation at the top of the list picks up exactly where we left off, recap and all. Any past conversation in the list can be reopened this way, not just the most recent one, so nothing gets lost if we step away or switch tasks partway through. KIBO AI also includes a separate “Workspace” area, that can open automatically for larger tasks that produce more than a short text reply, like a chart or a set of visuals. A request in the “Analytics” area asking to see revenue trends over time, for example, would generate its output there instead of inside the chat itself. KIBO AI also includes a separate “Workspace” area, that can open automatically for larger tasks that produce more than a short text reply, like a chart or a set of visuals. A request in the “Analytics” area asking to see revenue trends over time, for example, would generate its output there instead of inside the chat itself. We can expand the chat area as well, by clicking the “Outward Facing Arrows” icon. This helps when we need more space to read replies and context, or anytime more space is desired. We can click the reversed “Inward Facing Arrows” icon to revert back. We can expand the chat area as well, by clicking the “Outward Facing Arrows” icon. This helps when we need more space to read replies and context, or anytime more space is desired. We can click the reversed “Inward Facing Arrows” icon to revert back. Finally, clicking the “X” on the far right of the chat panel closes it completely. Finally, clicking the “X” on the far right of the chat panel closes it completely.
Want to learn more?
Kibo AI Overview
Learn more about this topic