Dapr Now Supports AI Agents | Distributed Application Runtime

Dapr Extends Capabilities to Support AI Agents

In 2019, Microsoft released Dapr as an open-source runtime designed to simplify the development of distributed, microservice-based applications. Interestingly, despite the absence of widespread discussion around AI agents at that time, Dapr inherently included foundational elements for their support.

Core Functionality: Virtual Actors

This capability stems from Dapr’s central concept of virtual actors. These actors can independently receive and process messages, functioning separately from other actors within the system. This inherent independence is crucial for agent-based systems.

Introducing Dapr Agents

The Dapr team is now introducing Dapr Agents, a dedicated solution aimed at assisting developers in building AI agents. It provides a comprehensive set of building blocks to streamline the development process.

Why Dapr is Well-Suited for Agents

Yaron Schneider, a Dapr co-creator and maintainer, highlights that agents are an ideal application for Dapr’s architecture. Actors offer a lightweight method for running agents at scale, maintaining state while remaining resource-efficient.

However, Schneider notes that significant business logic still requires implementation. While workflow engines or actor frameworks can handle statefulness and orchestration, substantial effort remains in crafting the agent logic itself. Dapr distinguishes itself by offering a superior level of orchestration and state management compared to many existing agent frameworks.

Origins in Floki

Dapr Agents evolved from Floki, an open-source project that initially extended Dapr for AI agent applications. Collaboration between the project maintainers and Microsoft AI researcher Roberto Rodriguez led to the integration of Floki under the Dapr umbrella, ensuring the framework’s continued development.

Agents as Distributed Systems

Mark Fussell, also a Dapr co-creator and maintainer, views agentic systems as a reinterpretation of distributed systems. He suggests that the term “agent” can be used interchangeably with “microservice,” particularly when incorporating large language models (LLMs) into the architecture.

Orchestration and Statefulness are Key

The team emphasizes the necessity of an orchestration engine and statefulness for effectively coordinating agents. Dapr provides both of these essential components.

Dapr’s actors are designed for exceptional efficiency, capable of initiating within milliseconds upon receiving a message and terminating—while preserving their state—once their task is completed.

Model Provider Integration

Currently, Dapr Agents offers out-of-the-box compatibility with leading model providers, including AWS Bedrock, OpenAI, Anthropic, Mistral, and Hugging Face. Support for local LLMs is scheduled for imminent release.

Tools and Task Fulfillment

Beyond model interaction, Dapr Agents, leveraging the existing Dapr framework, enables developers to define a suite of tools that agents can utilize to accomplish specific tasks.

Language Support

At present, Dapr Agents supports Python, with .NET support on the horizon. Further language support, including Java, JavaScript, and Go, will be added in the near future.