Asya
The Message Knows the Way.
/ΛΙΛ.sjΙ/ — async actors π
- A Kubernetes-native actor mesh for AI and agentic workloads
- Queue-based routing — each message carries its own route
- Independent scaling from zero to infinity
- No central orchestrator — choreography, not orchestration
- Pure Python functions and K8s manifests — no infrastructure in your code
Two files. Two owners.
Data Scientist owns the handler
Pure Python, zero dependencies. No SDK imports, no decorators, no framework coupling.
Test locally with assert process(state) == expected.
Platform team owns the spec
Deploy as a Kubernetes CRD. Asya injects the sidecar, creates the queue, configures KEDA autoscaling. Your code stays clean.
Why Asya
Built for AI/ML workloads that need independent scaling, fault isolation, and zero framework lock-in.
π Actor Mesh
Choreography on Kubernetes — no central orchestrator, each message carries its own route
βοΈ Message Knows the Way
Stateless actors, all state travels in the envelope — route, payload, headers
πΎ Virtual Actors
Stateless by design, with optional persistent memory via pluggable state proxy
βΈοΈ Kubernetes Native
CRDs, Helm charts, GitOps — runs anywhere K8s runs
π Scale Zero to Infinity
KEDA scales each actor independently — GPU pods cost nothing when idle
π Pluggable Transport
SQS, RabbitMQ, Pub/Sub — swap via Crossplane, handler code unchanged
ποΈ Pluggable Storage
S3, GCS, Redis, NATS KV — state proxy connectors for virtual actor memory
π‘οΈ Built-in Resiliency
Durable queues, retry policies, DLQ, SLA enforcement, timeout control
Architecture
Actors communicate through message queues. Each message is an envelope that carries its own routing instructions.
Actor Mesh: messages carry their own route through independently scaling actors
Actor Anatomy
Every actor pod has two containers: your Python runtime and the Asya sidecar. The sidecar polls the queue, delivers the envelope to your handler via Unix socket, and routes the response to the next queue in the chain.
Queue → Sidecar → Runtime → Sidecar → Next Queue
Quick Start
Local actor mesh in minutes.
Requires Docker, kubectl, Helm, and Kind.
Create cluster
Install Asya
Send a message
Interoperability
Pluggable at every layer. Swap transports, state backends, and gateway protocols independently.
How Asyaπ Compares
Honest, factual comparisons with real code examples.
π vs Agentic Frameworks
Distributed actors, not in-process graphs — LangGraph, CrewAI, ADK, AutoGen, KAgent
π vs Workflow Orchestrators
Choreography, not centralized orchestration — Temporal, Argo, Airflow, Prefect, Dagster
π vs ML Pipelines
Runtime-centric, not training-centric — KFP, Flyte, Metaflow, ZenML
π vs Stream Processing
Discrete messages, not infinite streams — Flink, Kafka Streams, Spark
π vs Actor Frameworks
Pure functions, not mailboxes — Erlang, Akka, Orleans, Dapr
All Comparisons
7 categories + 7 in-depth 1:1 with code examples