Build AI chatbots with Coze! All experience levels welcome.

The host, Zara, welcomes viewers to the live stream focused on building AI chatbots using the platform Co. She introduces Joshua, a developer advocate for Co, and they discuss the platform's capabilities. They mention that the platform is suitable for developers of all levels and highlight the bot store, which contains community-created bots for various purposes such as learning, productivity, and entertainment. Zara and Joshua also encourage viewers to ask questions and share their locations.

The live stream continues with a look at the bot store, where Joshua explains how users can find inspiration and even customize existing bots. They demonstrate how to interact with a GitHub expert bot, which uses plugins to access the GitHub API and browser functionalities. Joshua also shows how users can view the workspace that built the bot, including its persona, plugins, and skills, and how they can duplicate and modify the bot through public configuration.

Joshua guides viewers on how to start creating a chatbot, using a trip to Japan as an example. He outlines the process of naming the bot, choosing a large language model, and customizing its persona and skills through natural language prompts. The platform's ability to optimize prompts and use markdown formatting for better AI comprehension is also discussed. Joshua emphasizes the platform's features, such as knowledge bases, workflows, and multi-step processes, and how they can be leveraged to enhance the bot.

The live stream demonstrates testing the chatbot's capabilities by asking for restaurant recommendations in Ginza, Tokyo. The bot uses its built-in knowledge to provide an answer. Joshua then introduces plugins, which act as APIs to access various services and data. He shows how to add plugins like Google web search, DALL-E for image generation, and Trip Advisor to enhance the bot's functionality. The importance of specifying when to use certain plugins within the bot's skills is also highlighted.

Joshua explains the use of card bindings to structure bot responses, making them more organized and user-friendly. He shows how to use card bindings with the Yelp API to present restaurant recommendations in a visually appealing format. The process of selecting data fields from the API response and customizing the card's appearance is demonstrated. The feature of linking the card to an external page, such as a Yelp review, is also covered.

The live stream covers the use of variables to personalize the bot based on user preferences. Joshua creates a variable for trip preferences, which can store information about the user's interests, such as hiking or historical site visits. He shows how the bot can utilize this variable to provide more customized responses. The privacy of user-specific variables and the ability for developers to add more variables are also discussed.

Joshua introduces knowledge bases, which allow the bot to reference content from various sources like websites, Excel sheets, and APIs. He demonstrates how to create a knowledge base and add different units of knowledge, such as documentation pages, to enhance the bot's context and responses. The concept of retrieval augmented generation (RAG) and the customization of search strategies within knowledge bases are also explained.

The live stream delves into workflows, which combine various skills like plugins and knowledge bases to create multi-step processes. Joshua uses an NBA scores bot as an example to show how workflows can generate structured responses. He explains the use of nodes within workflows to pass information from one step to another, and how a code node can be used to parse data before it's processed by a large language model.

Joshua clarifies the difference between using plugins directly and incorporating them into workflows. He explains that while plugins provide access to various services and data, workflows allow for a more structured and predictable response. The live stream presents an example of generating images with a consistent style using DALL-E and a workflow, emphasizing the need for workflows when a plugin's output requires further processing.

The live stream covers the use of triggers to automate bot tasks based on specific events or schedules. Joshua demonstrates how to set up a trigger for the bot to send a message at a particular time without user interaction. He also shows how to use event triggers to respond to actions like receiving an email. The concept of allowing users to create their own triggers during a conversation is also discussed.

Joshua introduces databases in Co, which allow for the organization of data in a tabular structure. He demonstrates how to create a database table, add fields, and populate it with information through natural language interactions with the bot. The use of databases for tasks like contact list management is shown, and the distinction between single-user and multi-user modes for databases is explained.

The final part of the live stream focuses on publishing bots for wider use. Joshua explains the process of publishing a bot to the Co store and other chat applications like Discord and Telegram. He discusses the benefits of publishing to the bot store, including the potential for featuring in recommended sections and gaining exposure. The importance of sharing creations in community spaces, like the Co Discord server, is also highlighted.

The live stream concludes with a summary of the key points covered and an invitation for viewers to engage further with the platform. Joshua and Zara thank the viewers for joining, encourage them to explore the documentation, and join the Co Discord server for support and community interaction. They also mention that the session will be recorded and shared on YouTube and the dev Community for future reference.