Latest release cuts the data sent on each workflow step, adds an opt-in mode that removes a control-plane service and brings consistent workflow management APIs to the Dapr SDKs

Diagrid, developer of the world's most reliable foundation for mission-critical workflows and autonomous AI agents and the company behind the Distributed Application Runtime (Dapr), today announced the release of Dapr 1.19. The latest version of the Cloud Native Computing Foundation (CNCF)-graduated project makes workflows faster, lets operators run one fewer control-plane service and exposes workflow management APIs consistently across the Dapr SDKs.

AI agents run for hours, fan out into dozens of parallel tasks and must survive restarts without losing their place; durable workflows make that possible. A workflow's history also grows as it runs, and before 1.19, Dapr sent that entire history to the worker on every step. For long-running agents, costs added up quickly.

In Dapr 1.19, a workflow worker keeps the history it already holds and receives only new events. Previously, every step re-sent the full execution history, so long-running and high-fan-out workflows shipped a larger payload with each step. That pushed up latency, memory use and network traffic. Now, the amount of data sent per step no longer depends on how long the workflow has been running.

The worker still replays the full history locally; only the transfer shrinks. Workers opt in by advertising the capability, and Java and Python SDK workers support it in 1.19. Workers that don't advertise it keep receiving the full history.

A Faster Path Through the Workflow Engine

A new opt-in preview feature, WorkflowsFastPath, handles workflow wake-ups and local activities in place. It no longer routes every event through a durable scheduler reminder but keeps the durable recovery path as a backstop. In engine benchmarks, the feature reduced median workflow step latency from 553 milliseconds to 153 milliseconds and increased fan-out throughput. It is off by default in 1.19, so operators can adopt it on their own timeline.

In addition, stateful worker streams transmit only new history events (deltas) so the workflow engine doesn't need to ship the entire orchestration history every time.

"Agents made durable execution the default for production software, and they raised the bar on what it has to cost," said Mark Fussell, co-founder and CPO of Diagrid. "An agent that runs all afternoon and splits into 50 tasks can't haul its entire history across the network on every step. Dapr 1.19 keeps the work per step flat no matter how long a workflow has been running."

Dapr has shipped two control-plane services with overlapping jobs. The Placement service distributes the actor placement table. The Scheduler service handles jobs, reminders and workflow scheduling. Both track cluster membership, and both push state to sidecars.

Dapr 1.19 adds an opt-in mode in which the Scheduler also handles actor placement, so the standalone Placement service is no longer deployed. Teams that switch it on run one fewer stateful service and one fewer Raft cluster, with fewer failure modes to track. Reminders also get faster, because the Scheduler knows which sidecar owns each actor and delivers each reminder directly to it. Operators turn the mode on through a Helm setting once every sidecar runs 1.19 and can switch it on or off without restarting sidecars.

The new mode also makes scaling events less disruptive. When a host joins or leaves, Dapr now locks and updates only the affected actor types (it previously paused and resent the entire namespace's placement table). Actor types untouched by the change keep serving traffic. Placement also moves from a consistent-hash ring, which needs tuning to balance well, to rendezvous hashing, which balances optimally with no tuning and relocates the smallest possible set of actors when hosts change. During a rolling restart or scale event, more of the cluster keeps running undisturbed.

"A graduated CNCF project has to deliver releases teams can plan around, and 1.19 does that," said Yaron Schneider, co-founder and CTO of Diagrid and co-creator of Dapr. "Placement and Scheduler matured as two services doing overlapping work. Folding placement into the Scheduler gives the project a simpler control plane, which is where Dapr needs to go."

Additional Building Blocks for AI and Cloud-Native Applications

Dapr 1.19 also introduces new building blocks and .NET improvements to help developers build AI-powered and cloud-native applications:

• Vector building block (alpha): Provides a portable API for storing, retrieving and querying vector embeddings for AI applications, with filterable metadata and similarity search. Meilisearch is the first supported component, with more in development. A companion Search building block brings keyword search to the same model.
• Binary Store building block (alpha): Streams large files, including documents, images and model outputs, in chunks rather than loading entire files into memory. Initial components support AWS S3 and S3-compatible storage, Azure Blob Storage, Azure Data Lake Storage Gen2, Google Cloud Storage and Oracle Cloud Infrastructure Object Storage.
• Dapr Actors.Next for .NET: Introduces a new actor SDK implementation that generates proxies and registration at compile time, reducing runtime overhead and enabling Native AOT compatibility. A persistent connection to the Dapr sidecar and built-in analyzers further simplify development and catch common errors before production.

Availability

Dapr 1.19 is available today. Full release notes are available on GitHub, and additional details are covered in our blog post. Teams running Dapr on Kubernetes can manage the upgrade through Dapr Ops Dashboard from Diagrid, which typically supports new Dapr releases within 24 hours of public release.

About Diagrid

Diagrid helps enterprises build reliable and secure distributed systems and AI agents. Founded by the creators of the open-source Distributed Application Runtime (Dapr), Diagrid provides the agentic infrastructure layer that delivers durable execution, secure service‑to‑service communication, AI orchestration, and cryptographic trust for modern cloud‑native and AI applications. Its flagship product, Diagrid Catalyst, combines durable workflows with advanced identity, access controls, and zero‑trust style governance so teams can run long‑lived, mission‑critical agent workloads without losing state or compromising security. Diagrid is a leading maintainer of Dapr within the Cloud Native Computing Foundation (CNCF) and a member of the Agentic AI Foundation (AAIF).

Organizations in highly regulated sectors such as financial services and healthcare use Diagrid to orchestrate multi‑agent workflows, enforce fine‑grained data access, and cryptographically attest which agents did what and when across their infrastructure. Learn more at https://www.diagrid.io/.

View source version on businesswire.com: https://www.businesswire.com/news/home/20261109736602/en/

Latest Dapr 1.19 release cuts the data sent on each workflow step, adds an opt-in mode that removes a control-plane service and brings consistent workflow management APIs to the Dapr SDKs