Create an agent step by step
Last updated
Last updated
Here are the steps to follow to create an agent:
The instructions during the creation of an agent define its behavior, expertise, and limitations. They guide the agent on how to respond, which tone to adopt, and what information to provide. These personalized guidelines are essential for creating an AI assistant that is perfectly tailored to your specific needs.
In Swiftask, skills are specific features that you can add to your AI agents to extend their capabilities and allow them to interact with external tools or services. They function as add-on modules, enhancing the possibilities of your agent.
Long-term memory allows the AI agent to:
Remember the user's preferences and habits
Adapt its responses based on the conversation history
Provide more contextual and personalized answers
Avoid asking for information that the user has already provided
When your chatbot widget is live, all interactions are displayed directly in the Widget Inbox. This centralized view allows you to evaluate the chatbot’s activity and adjust its settings as needed. Acting as a 24/7 customer service, it archives the exchange history and delivers personalized responses using your knowledge base.
Within the agent creation interface on Swiftask, you can take it further.
Here, you have the widget option. It is used to configure the chatbot to be placed on your website. You can set:
its activation,
the number of messages,
file sending,
the agent's name,
the subtitle to be displayed in the chatbot,
as well as the placeholder and the welcome message.
Below is the management of the agent's avatar, its color, its position on the site, and the shape of the icon to display.
To help the user communicate better, you have the option to add startup questionnaires—which is quite practical at the moment!
And at the bottom of the page, you have the integration codes. Open them and add them to your site using the method that suits you.
Access here the complete logs of your agent, including exchanged messages, encountered errors, warnings, and debugging information.
Logs are essential for:
Tracking interactions: They document all conversations between users and your AI agent, allowing you to review the history of exchanges.
Debugging and improving: They help you identify problems in the agent's responses or situations where it didn't correctly interpret requests.
Analyzing performance: They allow you to evaluate your agent's effectiveness by studying the types of questions received and the relevance of the answers provided.
Ensuring compliance and audit: They maintain a verifiable record of interactions, particularly important in sectors subject to strict regulations.
This functionality is essential for several reasons:
Verifying the quality of responses: It allows you to test how your agent responds to different questions or scenarios to ensure that it provides accurate and relevant information.
Identifying weaknesses: By testing your agent with various queries, you can pinpoint areas where it lacks knowledge or responds inappropriately.
Optimizing instructions: The results of evaluations help you refine your agent’s instructions and parameters to enhance its performance.
Validation before deployment: This step ensures that your agent is ready to interact with real users and accurately represents your brand or service.
Comparing versions: You can assess different configurations of your agent to determine which one best meets your specific needs.
This section appears once you've configured your agent and allows you to technically leverage your agent in several ways.
Widget Integration:
This option provides you with the necessary code to integrate your agent as a widget on your website or intranet.
You can simply copy/paste the provided HTML/JavaScript code into your website without needing advanced programming skills.
Iframe Integration:
An alternative to widget integration, allowing you to embed your agent in an iframe on your site.
This method can be useful for isolating the agent's interface from the rest of your website.
API Integration:
This section provides information for programmatically interacting with your agent via APIs.
It includes endpoints for sending messages, receiving responses, and managing user sessions.
Particularly useful for developers who want to create custom interactions or automate tasks.
Agent Identifiers:
Agent Client Token: A unique identifier for your agent used for authentication when making requests to Swiftask's backend services.
Agent Slug: A user-friendly identifier for your agent, typically used in URLs and API calls.
Conversation Starters Customize the beginning of the conversation to provide a tailored welcome.
Temperature Adjusts the bot’s creativity: high values result in varied but unpredictable responses, while low values produce more consistent responses.
Top K chunk Defines how many parts of documents the system retrieves from the indexed ones. This helps limit the number of results and speeds up the search.