Diagrid Inc. has released Dapr 1.19, introducing updates to its workflow engine designed to reduce latency and memory usage for long-running AI agents. The update changes how execution history is handled, sending only new events to workers instead of the full history. This feature is currently supported for workers built with Java and Python SDKs.
The new WorkflowsFastPath feature, available as an opt-in preview, processes local activities and wake-ups in place. According to company benchmarks, this change reduced median step latency from 553 milliseconds to 153 milliseconds. The feature uses the existing reminder path as a recovery mechanism.
Dapr 1.19 introduces a new mode that consolidates control-plane services by retiring the Placement service. The Scheduler service now manages actor placement, eliminating the need for a separate stateful service and its associated Raft consensus cluster. This update also improves scaling by locking only affected actor types during host changes.
New alpha building blocks include a Vector component for similarity search with Meilisearch support and a Binary Store for streaming large files to cloud storage like Amazon S3, Microsoft Azure, and Google Cloud. Additionally, .NET developers receive the Actors.Next SDK, which enables compile-time proxy generation for reduced runtime overhead.
Dapr 1.19 is available now for Kubernetes users. The Scheduler placement mode can be enabled via Helm settings without requiring sidecar restarts. Teams can manage the upgrade process using the Diagrid Dapr Ops Dashboard, which typically supports new releases within 24 hours of launch.
