Agents
AI agents are applications that use LLMs to interact with data available in their environment and automatically reason and take actions. Using agentic workflows, users can build applications that decompose problems, prioritize steps, and ultimately execute actions that will help achieve clearly defined objectives. This can enhance the ability of GenAI, by enabling the automation of tasks that require some decision making and the calling of external or custom logic or tools.
Our platform provides key functionality to support the development of effective agentic applications:
Tool Use empowers agents to extend their capabilities by interfacing with and calling external systems and APIs. Our Tool Use framework provides structured ways for users to deploy functionality that an agentic LLM can easily route to for execution based on it's decision-making as it aims to achieve the goal defined for it.
Planning capabilities enable agents to approach problems strategically by organizing work into coherent sequences of actions. Our planning functionality helps agents decompose tasks, prioritize steps, and establish clear objectives. With Seldon's support for planning workflows, deployed agents can go loop through different possible options iteratively as part of their decision-making process in order to automatically resolve a task.
[Reflection] (coming soon) mechanisms allow agents to evaluate their own performance, learn from experiences, and improve over time.
In this section you will find implementation guidelines, examples, and best practices to help build powerful agentic applications that leverage the patterns outlined above.
Last updated
Was this helpful?