AI
Last updated
Last updated
On BounceClub AI, Agents are the primary entities providing AI services. Each Agent can process end-user requests based on specified strategies, knowledge bases, preferences, and base models. An Agent is not just a simple model invoker but an intelligent entity capable of task handling and feature presentation. By integrating with Knowledge Bases, Agents can provide specialized and customized AI services.
Guide:
In the built-in Agent module, Agents are divided into two layers: the Access Layer and the Implementation Layer.
Access Layer: This is the main body of the Agent, containing the Agent's basic information, unique ID, knowledge base permissions, fee account, etc. The Access Layer is where users manage, distribute, and update the Agent. The information in the Access Layer does not affect the Agent's implementation. The design philosophy is to decouple the Agent's carrier from its core, allowing users to freely select, update, and modify Agents.
Implementation Layer: This is the core of the Agent, affecting the Agent's performance. This layer can be implemented using Agent frameworks (such as langchain) or Agent platforms (such as GPTs, Character.AI, Coze). For users without development experience or those who want a quick experience, BounceClub includes a lightweight Agent framework called "MiniCharacter," which allows users to create a high-performance Agent through simple form filling and supports a certain degree of customization and development in the future.
Why Access Agents Instead of Models Directly
Direct access to models can provide basic AI functionality but cannot meet complex and diverse application needs. Agents build on this by adding task management, strategy selection, and result optimization functions, making AI services more intelligent and efficient. Additionally, Agents can manage and coordinate the invocation of multiple models, providing a unified service interface, simplifying user operations and integration complexity.
Task-Oriented or Feature-Oriented Agents
Agents are usually oriented towards specific tasks or features and require significant code development. Agent Providers are the most direct contributors to improving end-user experience by writing and optimizing Agent code to make AI services more aligned with user needs and scenario requirements. For example, a customer service Agent needs capabilities in natural language processing, sentiment analysis, and problem-solving, while an academic research Agent needs strong literature retrieval and data analysis capabilities. The work of an Agent Provider includes designing and developing processes, managing modal data, and utilizing knowledge bases.
The Role of Usage in Agents
Usage plays a critical role in the management and optimization of Agents:
Upgrading Agents: By collecting and analyzing Usage data, Providers can identify performance bottlenecks and improvement points in Agents, allowing for targeted upgrades and optimizations.
Collecting Feedback: Usage data includes user behavior and feedback information, helping Providers understand user needs and experiences and make corresponding adjustments.
Optimizing Aggregator Strategies: When developing Agents, AI services might use aggregator strategies, such as AI model aggregators and knowledge base aggregators. The Agent will then automatically analyze which AI service to use. Accumulating Usage data can help optimize aggregator strategies, selecting the most suitable models and service paths, thereby improving overall service quality and efficiency.