DAPR microservices now support AI agents
Back in 2019, Microsoft DAPR Open SourceA new execution time to easier facilitate applications based on distributed microservices. At the time, no one was still talking about AI agents, but ultimately, DAPR had some of the fundamental constituent elements to support the agents of the integrated AI from the start. This is because one of the main characteristics of DAPR is a virtual concept actorsWho can receive and process messages, regardless of all other players in the system.
Today, the DAPR team launches DAPR agents, its maintenance to help developers build AI agents by providing them with many construction blocks to do so.
“Agents are a very good use for DAPR,” said the co-creator and DAPR manager Yaron Schneider. “From a technical point of view, you can use the actors as a very slight way to execute these agents and really to execute them on a large scale with a state – and to be resource economics. All this is great, but then there is still a lot of commercial logic that you must write. The state and its orchestration are only part. And many people, they could choose a workflow engine or an actor executive, but there is still a lot of work to really write the logic of the agent on the other side. There are many agent executives, but they do not have the same level of orchestration and state as DAPR. »»
DAPR agents are from HerdA popular opening project that extended the DAPR for this case of use of the AI agent. By discussing the project managers, including Microsoft AI researcher, Roberto Rodriguez, the two teams decided to put the project under the DAPR umbrella to ensure the continuity of the new agent executive.
“In many ways, we see aging systems and all the terminology around that like any other term for” distributed systems, “said the co-creator and maintenance of the DAPR, Mark Fussell. “”[…] Rather than calling them microservices, you can call them agents now, mainly because you can put large language models among them all. »»
To effectively coordinate these agents, you need an orchestration engine and state state, supports the team – which is exactly what DAPR offers. This is partly because DAPR actors are supposed to be extremely effective and capable of shooting in milliseconds when a message arrives (and firm, with their preserved state, when their work is finished).
Currently, DAPR agents can speak to most suppliers of popular models outside the box. These include the AWS base, OpenAI, Anthropic, Mistral and Embringe Face. The care of local LLMs will arrive very soon.
In addition to interacting with these models, since DAPR agents extend the existing DAPR frame, developers also have the possibility of defining a tool list that the agent can then use to accomplish a given task.
Currently, the DAPR agents support Python, with the .NET support which is launching soon. Java, Javascript and Go will follow soon.
