Akka, known for its tools for building resilient and elastic distributed systems, has launched new deployment options designed to simplify the deployment of large-scale agentic AI systems. This announcement aims to provide developers greater flexibility in deploying systems across various infrastructure choices.
Companies like Capital One, John Deere, Tubi, Walmart, and Swiggy use Akka’s tools. The company now offers enterprises more freedom to deploy Akka applications on their preferred infrastructure. The new options include self-managed Akka nodes and self-hosted Akka Platform regions.
In an Akka blog post, Tyler Jewell, CEO of the company, stated:
Agentic AI has become a priority with enterprises everywhere as a new model that can potentially replace enterprise software as we understand it today. With today’s announcement, we’re making it easy for customers to build their distributed systems, including agentic AI systems, without having to commit to the Akka Platform. Now, enterprise teams can quickly build scalable systems locally and run them on any infrastructure they want.
Akka highlights that shifting towards agentic architectures requires a fundamental change from transaction-centered to conversation-centered systems. Agentic services maintain state within the service itself, storing each event to track how the service reached its current state, contrasting traditional stateless SaaS applications. Moreover, this shift introduces challenges for developers, including unpredictable behavior, limited planning, memory-impacting agent effectiveness, hard failures at scale, opaque decision-making, and increased costs and latency.
To address these challenges, Akka has introduced two new deployment capabilities:
- Self-managed Akka nodes: Developers can now run clusters of services built with the Akka SDK on any cloud infrastructure. The new version of the Akka SDK includes a self-managed build option that creates services that can be executed standalone. These services are packaged as binaries in Docker images and can be deployed in any container PaaS, bare metal hardware, VMs, edge nodes, or Kubernetes, without requiring Akka Platform dependencies. Akka clustering is built-in.
- Self-hosted Akka Platform regions: Teams can now run their own Akka Platform region without relying on Akka.io control planes. While services built with the Akka SDK have always been deployable onto the Akka Platform, Akka previously provided managed services through its Akka Serverless and Akka BYOC offerings. Self-hosted regions are Akka Platform regions with no Akka control plane dependency, which teams install, maintain, and manage. Akka will cooperate with customer SRE teams to install self-hosted regions.
Akka emphasizes that these new options provide greater deployment flexibility than other frameworks, particularly for agentic AI.
InfoQ spoke with Akka’s CEO, Tyler Jewell, and asked, “How do these new deployment options, particularly self-managed Akka nodes, address the specific challenges of deploying stateful agentic AI applications at scale, especially regarding data management, consistency, and fault tolerance?” Jewell answered that:
Akka is the only development framework offering built-in clustering, whether offline or on the Akka Platform. Akka services can form their clusters through discovery, allowing both stateful and stateless services to be inherently clustered. While you can operate a single-node cluster, a multi-node cluster is easily achievable. Clustered services benefit from features like discovery, data sharding, and brokerless messaging, enhancing their elasticity and resilience. With over 15 years of production experience, Akka’s clustering is embedded within services, eliminating the need for additional infrastructure.
In addition, InfoQ asked, “beyond infrastructure flexibility, what other features or capabilities does Akka provide to ensure the trustworthiness and observability of agentic AI systems deployed using these new methods?”. With Jewell explaining:
Agentic AI Platforms differ significantly from traditional cloud-native development frameworks. They enable the design, development, and operation of distributed systems composed of multiple agents that can transact and reason stochastically.
Key features of effective Agentic AI Platforms include:
1. High development velocity, allowing agents to scale, recover, and update efficiently.
2. Evaluation readiness, ensuring that reasoning processes are auditable, visible, measurable, and suitable for experimentation.
Evaluation in Agentic AI is ongoing, requiring continuous monitoring and operational strategies integrating DevOps and EvalOps practices. This includes visualizing reasoning, auditing decision flows, and employing multi-model testing.
Traditional microservices platforms, like Spring or Temporal, may facilitate the creation of deterministic systems but do not accommodate the uncertainty inherent in Agentic AI systems. Akka is emerging as a robust Agentic AI platform, supporting:
- Collaboration among numerous agents across various networks with durable execution and zero-trust networking.
- Integration of Domain-Driven Design (DDD) with AI to enhance deterministic systems.
- A comprehensive evaluation strategy that ensures reasoning is visible and measurable throughout the system’s lifecycle.
- Enable resilient, elastic execution of agentic services that have AI with data at rest and data in motion; failover and DR for memory; SDK and tooling to simplify the creation and testing of multi-agent systems; and native support for all tooling, MCP, and A2A protocols.
