AI Gateway
Overview
The AI Gateway serves as the central hub for creating, configuring, and deploying AI agents across your platform. It lets you design intelligent assistants that can understand context, access knowledge, and interact naturally with users — all without complex setup.
Each agent you create can have its own personality, knowledge base, and behavior settings. You can train it with your own data, adjust its tone and formality, and integrate it seamlessly into your web applications through a simple script or standalone mode.
A typical setup flow looks like this:
- Create the agent — Start by defining a new AI assistant.
- Describe the role — Define its purpose and personality traits.
- Set up the knowledge base — Train the agent with documents, links, or Q&A data.
- Configure behavior — Fine-tune models, tone, and safety settings.
- Integrate the agent — Embed or launch it in your application.
Agent creation
Multiple bots can be created. Each with its own persona, knowledge base and configurations. Click on "Add Agent" to setup a new chatbot.
Bots can be deleted from this page by clicking on delete option on the three dot menu.
Describe the agent
The user must define in the description the purpose of the AI agent. There are some pre-defined options provided for the user to select. If a user wishes to define their own role they can. Once the agent is created, users will be taken to the AI Persona page where users can configure the personality of the AI. The tags will be auto generated based on the description provided in the "Describe the agent" section.
Persona
This section is where the photo of the AI can be changed. The name of the bot can also be defined here in the "Agent Name" field.
Some behavior settings are here:
Tone - Controls how the AI shapes its replies. Ranages from concise to conversational.
Formality - Adjusts how fomal or balanced the AI sounds.
Technical Depth - Controls how deeply the AI explains technical concepts.
Setting up the Knowledge Base
It is recommended to add the knowledge base of the model. This will help keep your model’s responses within the context of your app. You can train it with four types of data.
Knowledge can be fed in the following ways:
- Knowledge – Raw text can be inserted here. This editor also supports markdown.
- File – currently supports .txt, .pdf, Microsoft word file. At a time up to five files can be uploaded. There is a limit of 5MB per file.
- Link – links to resources. There is an option to periodically crawl the site. There is also an option to enable recursive crawling so that the links within the page will be followed and crawled.
- Question & Answer - Simple question and answer format input for responding to specific questions.
To change the name of the bot, head over to the global section and edit the Name field.
The uploaded knowledge can be previewed by clicking on the Preview option in the three dot menu.
When you change the embedding model, knowledge needs to be re-embedded.
Behavior Settings
You must click save at the bottom of the page for the changes to take affect.
Some of the settings below appear only when you enable "Show advance settings"
This section gives users deep control over the behavior of the bots. Specially useful for technical users who have working knowledge of RAG models. The following are categorized sections for modifying the chatbot's behavior:
LLM Provider Configuration
Select Model - Allows selection of the model. Currently available options are
- GPT-4o Mini
- GPT-4o
- GPT-5
Instruction set
- Goal - Set the AI's primary objective. Guides decision - making and keeps respnsese aligned with its intended purpose.
- Backstory - Provides fictional or factual background. Helps the AI stay in character and reason consistently based on its identity.
Safety and moderation
Guardrail Instructions - Sets high-level business rules or limits the AI must follow. e.g., “Always be positive about our company’s products.”
Pre Guardrails - Rules applied before sending user input to the model (e.g., filtering or modifying prompts).
Post Guardrails - Rules applied after the model generates a response (e.g., filtering or adjusting output).
Banned Keywords - List of words or phrases that will automatically trigger a block if detected.
Violation Message - Custom message shown to users when their request is blocked by guardrails.
Conversation Behavior
FAQ Mode - Enable predefined common questions for quick answers.
Smart Suggestions - Allow the agent to suggest helpful prompts during conversation.
Manage FAQ Questions - List of common questions the agent can answer (maximum of 5).
Playground
Click on the playground button to interact with the bot. Get a feel for how the bot behavior without leaving the platform. It is helpful to have it here so you can tinker with the settings and see the changes happen.
It is possible test the bot in a standalone version as well. See Standalone mode.
See the source and conversation history
Go to the conversations page by clicking on the Conversations button on the top right. Here you will be able to see the past conversations along with the sources from which the information was taken.
Integration
Integration can be done in the following ways:
Embed the script
In order to use the chatbot in your front-end app, go to the publish tab under the design section, copy the widget script, paste it in your front-end code. The code will look like this:
<script
src="https://gpt.seliseblocks.com/embed.js"
data-widget-id="dc5***************************ae"
data-widget-type="chat"
data-project-key="18E***************************7C"
data-app-domain="https://p******.seliseblocks.com"
data-app-mode="dev">
</script>
Typically, you would paste the script at the bottom of your page right before the </body> tag
Standalone Mode
An immersive view of the chatbot. Copy and paste the link generted in the publish tab into a new tab or window or open it directly by clicking on the open button.
Scan the QR code
Scan the QR code and you will be taken to a immservive view of the chatbot.
Customize the look and feel
Go the "Design" tab in the Integration page.
💬 Chat Widget Customization
This screen allows you to customize the look and feel of your chat assistant. Any changes you make on the right-hand panel will be instantly reflected in the live preview on the left side, so you can see your design in real time before saving.
Layout Overview
- Left Side: Live preview of your chat widget.
- Right Side: Configuration panel with styling and color options.
- Save Button: Applies and stores all your customization changes.
Customization Options
Chat Header
Control how the top bar of your chat widget appears.
- Start Color / End Color: Define the gradient colors for your chat header background. The gradient flows from Start to End color.
- Title Color: Sets the color of the chat header title (e.g., “CareLink”).
Chat Style
Customize the main chat area where messages appear.
- Background Color: Changes the background of the chat window.
- Font Color: Defines the color of the text inside chat messages.
- Button Color: Controls the color of the message input area and action buttons.
- Icon Color: Sets the color of icons such as the send arrow or attachment icons.
- Border Color: Changes the color of the chat window’s outer border and message box outlines.
Toggle Button
Adjust how the floating chat toggle button (used to open or minimize the chat) looks.
- Background Color: Sets the color of the chat toggle button.
- Icon Color: Changes the color of the chat icon inside the toggle button.
Save Changes
Once you’re satisfied with your customizations, click Save to apply them permanently. Your chat widget will now reflect the new style settings across your platform.
You can automatically populate the stylings by adding a link to a site. The bot will crawl the site and learn the stylings and apply them to your agne widgets. Look for the "Site Url" field in the Design Tab.
Change the Greeting
Go to Integration. Click on the Global tab. Set the message here.