Dapr’s microservices runtime now supports AI agents
In 2019, Microsoft Open-Sourced Dapr, a new runtime for making buildings easier on micrenservice-based applications. At the time, nobody was talking about AI agents, but it turned out that DAPR had some of the fundamental building blocks for supporting AI agents who have been built in from the start. That’s because one of the core characteristics of DAPR is a concept of virtual actorsThose messages can receive and process, regardless of all other actors in the system.
Nowadays, the DAPR team launches DAPR agents, the view of helping developers to build AI agents by offering them a lot of the building blocks to do this.
“Agents are a very good use case for DAPR,” explained Dapr maker and subordinate Yaron Schneider. “From a technical perspective, you could use actors as a very lightweight way to run these agents and to be able to run them on a scale with the state and to be a resource efficient. This is all great, but then there is still a lot of business logic that you have to write. The state and orchestration are just one part. And many people can opt for a workflow -engine or an actor framework, but there is still a lot of work that they have to do to write the agent logic on the other side. There are many agent frameworks that are there, but they do not have the same level of orchestration and state that Dapr has. “
DAPR agents come from FlokiA popular open-source project that DAPR has expanded for this use case of AI agent. In conversation with the project researchers, including Microsoft AI researcher Roberto Rodriguez, the two teams decided to bring the project under the DAPR paraplu to guarantee the continuity of the new agent framework.
“In many ways we see agentic systems and the entire terminology that as a different term for” distributed systems, “said Dapr-Maker and submerged Mark Fussell. ‘[…] Instead of calling them micro services, you can now call them agents, especially because you can place large language models among them all. “
To efficiently coordinate those agents, you need an orchestration engine and is needed, the team argues – that is exactly what DAPR delivers. This is partly because the actors of DAPR are intended as extremely efficient and able to run within milliseconds when a message enters (and closing, with their state stored when their work is done).
At the moment DAPR agents can talk from the box with the most popular model providers. These include AWS rock, openi, anthropic, mistral and cuddly face. Support for local LLMS is coming very soon.
In addition to interaction with these models, because DAPR agents expand the existing DAPR framework, developers also have the option to define a list with tools that the agent can then use to perform a certain task.
DAPR agents are currently supporting Python, with .NET support that will be launched soon. Java, JavaScript and Go will follow soon.
