Behavior · State · Endpoints

Akka SDK

Akka SDK is the component model for building agentic systems.

Scroll

An SDK that models entire AI systems.

Akka components compose into agentic services that have their own APIs, orchestration, durability, and resilience so that an agent that fails mid-task resumes rather than restarts.

Otherwise you assemble
Agent frameworkOrchestration engineMessage brokerMemory / vector storeJob schedulerAPI frameworkState databaseCache
▸
Akka SDK

One programming model. Every component is a clustered microservice, so the infrastructure you would otherwise integrate is already built in.

clusteringdurable statestreamingschedulingendpoints

Create agentic systems from composable parts.

YOUR SERVICE AKKA RUNTIME AI models MCP clients Clients Message broker tools call read step delegate events · state changes topics TOOLSMCP endpoint JSONHTTP endpoint PROTOBUFgRPC endpoint SESSION MEMORYAgent MULTI-STEPWorkflow AT LEAST ONCETimed action ACROSS ENTITIESView DURABLE PROCESSAutonomous agent EVENT JOURNALEvent sourced entity CURRENT VALUEKey value entity EVENT STREAMConsumer
Component client

The component client is how one component calls another.

Agents

An agent interacts with an AI model to perform a specific task.

Autonomous agents

Agent and task state persist as it goes, so work survives crashes and restarts.

Workflows

A business transaction lives in one place, and the workflow either keeps it moving or rolls it back when a step fails.

Event sourced entities

State is rebuilt by replaying those events, and gives you a complete history of how the state came to be.

Key value entities

Akka guarantees exactly one instance of each entity across the whole cluster, so commands are handled one at a time.

Views

Views let you read across many entities, or find an entity by something other than its id.

HTTP endpoints

HTTP endpoints accept and return JSON by default.

gRPC endpoints

Protobuf’s forward and backward compatibility make service-to-service calls safe to evolve without breaking existing clients.

MCP endpoints

What you expose becomes tools the model can call on its own behalf.

Timed actions

Timers run at least once: if the call fails, the timer reschedules itself until it succeeds.

Consumers

Consumers read a stream of events, from an entity’s journal, from key value state changes, or from a message broker topic.

We’ve been building distributed systems since 2009.

Services built with Akka have embedded clustering to make your AI resilient from failures.

in-memory actor-based concurrency & clustering
Hot processes that run in memory, replicate across the cluster, and survive any node.
in-memory stateful, durable & sharded data
Durable in-memory state that's sharded across the cluster and replayable from its own event journal.
in-memory real-time event-driven streaming with backpressure
Continuous, backpressured event flows across components, services, and regions, without an external broker.

If it compiles, it's production-ready.

You write the domain logic. Resilience and scalability are built into the runtime, so the system that runs on your laptop is the system that runs in production.

YOUR LAPTOPYOUR SERVICEDomain logicComponent behaviorBusiness rulesAKKA RUNTIME✓ COMPILESSAME SYSTEMPRODUCTIONREPLICATINGREGION ANODE 1Your serviceNODE 2Your serviceNODE 3Your serviceNODE DOWNREGION BNODE 1Your serviceNODE 2Your serviceNODE 3Your serviceSCALED TO ZERORESTARTRESUMEDClustering & shardingClustering & shardingIn-memory durable stateIn-memory durable stateFailover & replicationFailover & replicationActive-active HA/DRActive-active HA/DRElastic scale-to-zeroElastic scale-to-zeroDurable executionDurable executionWHAT THE RUNTIME GUARANTEES

An entire AI system in one binary.

Agents, controls and model routing compile with the Akka runtime into one service, and its replicas form a cluster without a service mesh.

One service binary
Agents and workflows · memory · tools
Controls · guardrails · sanitizers · evals · human approval
Model routing
Endpoints · HTTP · gRPC · MCP
Akka runtime
Replicas self-cluster

StartEach replica owns a share of the agent sessions.

ScaleA new replica joins and takes its share.

×

RecoverA replica fails and its sessions resume on the others.

An integrated platform.

The platform uses smart shared abstractions to make every AI system as efficient as possible.

Akka Specify builds agentic systems from your specifications and releases only what meets your definition of done.→
Akka Verify applies any control to any AI system and guarantees its enforcement.→
Akka Optimize puts each request on the best model for the task, including smaller models trained on your data.→
Akka SDK

Everything an agentic system needs.

Behavior, state, and endpoints in one model.

Get started →