DigitalOcean GenAI Platform lets you build GPU-powered AI agents with fully-managed deployment. Agents can use pre-built or custom foundation models, incorporate function and agent routes, and implement RAG pipelines with knowledge bases.
The GenAI Platform includes the following key features:
-
GenAI Agents: Build agents using foundation models, including testing and comparing various foundation models in the Model Playground.
-
RAG Pipelines and Knowledge Bases: Implement RAG pipelines, a method that retrieves and integrates information beyond training data by creating and managing knowledge bases for more accurate and context-aware responses.
-
Agent Routing: Set up multi-agent experiences with agent routing to ensure the right agent handles each task.
-
Code Snippets for Chatbots: Integrate your agents into chatbots with code snippets.
Models
Agents
- Agent Endpoint: Ensures your agents are available for internal resources and external applications, either through a provided chatbot or by integrating the endpoint into their own app. If an endpoint is set to private, this restricts access to internal resources only, such as for agent routing, or you can use a key to allow internal users to access the chatbot. If an endpoint is set to public, the endpoint becomes accessible to external traffic. The endpoint enables you to use the OpenAI spec.
- Agent Routing: Directs tasks or queries to the most suitable specialized agent based on the nature of the input, ensuring the query is handled by the agent best equipped to respond.
- Agent Playground: A testing environment where you can experience the full functionality of your configured agent, including interactions with Knowledge Bases (KBs) and agent routing. While in the Agent Playground, you can adjust the agent’s model and attached KBs. You can use this tool to evaluate your agent’s behavior at any time after it is created.
- Example Agents: These agents model best practices for agent design and how retrieval-augmented generation works with a knowledge base. You can also use these example agents as a starting point for your custom agent.
Knowledge Bases
- Knowledge Bases (KB): A privately stored repository of unstructured files, folders, or web pages that enhance agent responses through Retrieval Augmented Generation (RAG). By attaching a KB to an agent, the agent can incorporate relevant information from the KB, even if it wasn’t originally known by the foundation model. The KB can include documentation, FAQs, and guides, enabling efficient retrieval of contextually relevant information for more accurate responses.
- OpenSearch, an open-source search and analytics tool for analyzing and working with data, is used to create Knowledge Base (KB) storage solutions. This means that when you create a KB, it requires a new or existing OpenSearch database cluster. For more information about OpenSearch, see OpenSearch documentation.
Functions
- Function Routing: Allows an agent to trigger predefined functions or APIs based on specific user inputs or tasks. This allows the agent to retrieve information from application databases or external APIs. For more information, see How to Route Functions in Your Agent.
Guardrails
- Guardrails: Overrides an agent’s output when it detects sensitive or inappropriate content in the input or output. For example, guardrails can prevent an agent from sharing login credentials or credit card information.