Diagrid Updates Dapr Runtime for the Agentic AI Era

Diagrid at the KubeCon + CloudNativeCon North America 2026 conference has made available an update to the open source Distributed Application Runtime (Dapr) that makes it possible to deploy artificial intelligence (AI) agents in a way that consumes infrastructure resources more efficiently.

The Dapr framework is based on Actors that are used to build stateful, concurrent, and fault-tolerant distributed objects, while Workers execute discrete tasks assigned by the workflow engine.

Previously, Dapr would send the full execution history of a workflow to a worker for each execution. Version 1.19 of Dapr now makes it possible for a worker to retain the history of a workflow, which means only updates to the history of a workflow now need to be shared.

Additionally, there is now an opt-in preview feature, dubbed WorkflowsFastPath, that handles workflow wake-ups and local activities without routing every event through a durable scheduler reminder. At the same time, a stateful worker now only transmits the history of new events back to the workflow engine.

Historically, Dapr made use of two control-plane services. A Placement service distributes the actor placement table, while the Scheduler service handles jobs, reminders and workflow scheduling. Dapr 1.19 adds an opt-in mode in which the Scheduler also handles actor placement to eliminate the need for a separate Placement service.

Dapr also now locks and updates only the affected actor types. Placement also moves from a consistent-hash ring to rendezvous hashing to eliminate the need for tuning.

Diagrid is also making available in alpha an application programming interface (API) for storing, retrieving and querying vector embeddings for AI applications and a Binary Store that makes it possible to stream large files.

Finally, there’s a new software development kit for .NET environments that generates proxies and registration at compile time. Originally developed by Microsoft, Dapr was donated to the Cloud Native Computing Foundation (CNCF) in 2021 and officially became a Graduated project in 2024.

Mark Fussell, chief product officer for Diagrid, said as application development teams move to deploy artificial intelligence (AI) agents there is a growing need for a runtime that is more resilient. Dapr in the event of a disruption makes it possible to recover a workflow versus requiring an AI agent to restart it from the beginning, he added. That’s critical because AI agents can run for parallel tasks for hours that, in the event of a need for a restart, can be recovered by Dapr, said Fussell.

Just as importantly, Dapr also helps ensure that infrastructure resources are being consumed more cost effectively as potentially thousands of AI agents are deployed, he noted.

It’s not clear to what degree DevOps teams might need to adopt new frameworks for building software in the agentic AI era, but the one thing that is certain is any and all options are now on the table. The challenge, as always, is marshalling all the resources required to adopt a new framework when there are already so many that have been deployed to support any number of existing legacy applications.