In 2019, Microsoft released Dapr as an open-source project designed to simplify the creation of distributed microservice applications. At that time, discussions around AI agents were minimal, yet Dapr already included essential components that facilitate the development of AI agents. This is largely due to one of Dapr’s key features: the virtual actors concept, which allows these actors to receive and process messages independently of one another within the system.
Today, the Dapr team is unveiling Dapr Agents, a new initiative geared towards aiding developers in the creation of AI agents by offering a wealth of foundational elements needed for this task.
“Agents represent an excellent application for Dapr,” noted Yaron Schneider, co-creator and maintainer of Dapr. “Technically speaking, actors can serve as a highly efficient means to execute these agents, enabling scaled operations with state management while optimizing resource use. This is beneficial, but developers must still write considerable business logic. Statefulness and orchestration are merely part of the puzzle. While many might opt for a workflow engine or actor framework, significant work remains to be done in formulating the agent’s logic. Numerous agent frameworks exist, but few offer the same orchestration and state management capabilities as Dapr.”
Dapr Agents emerged from Floki, a well-regarded open-source initiative that adapted Dapr for AI agent applications. After discussions with project maintainers, including Roberto Rodriguez, an AI researcher at Microsoft, the decision was made to integrate Floki into the Dapr ecosystem to ensure the longevity of the agent framework.
“In many respects, we regard agentic systems and the entire concept surrounding them as synonymous with ‘distributed systems,’” stated Mark Fussell, co-creator and maintainer of Dapr. “[…] Instead of labeling them as microservices, we can now refer to them as agents, particularly since large language models can be integrated among them.”
To effectively manage these agents, the Dapr team emphasizes the necessity for an orchestration engine and state management—exactly what Dapr offers. The efficiency of Dapr’s actors allows them to be initialized within milliseconds upon receiving a message, and they can cease operations while preserving their state once their tasks are completed.
Currently, Dapr Agents are compatible with a variety of leading model providers right out of the box, including AWS Bedrock, OpenAI, Anthropic, Mistral, and Hugging Face. Support for local LLMs is on the horizon.
In addition to interacting with these models, Dapr Agents enhance the existing Dapr framework by allowing developers to specify a range of tools that the agents can leverage for task completion.
At present, Dapr Agents support Python, with upcoming support for .NET. Java, JavaScript, and Go will be added shortly.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
