Introduction

Athyr is a runtime for AI agents as distributed microservices.

Most AI agent frameworks run multiple agents as objects in the same process. That works for demos. In production, when agents need to run on different machines, scale independently, and communicate across networks, those frameworks break down.

Athyr treats agents as network services, not library objects. The platform handles the infrastructure—memory, messaging, LLM routing, state—so agents contain only business logic.

The Problem

Multi-agent systems built as same-process Python objects face hard limits:

In-Process AssumptionProduction Reality
Shared memoryNetwork serialization
Instant function callsNetwork latency (ms to seconds)
Single failure domainPartial failures, timeouts
Exceptions propagateMust handle disconnects

When agents become independent services across machines, you need infrastructure designed for distribution from the start.

What Athyr Provides

Deployment & Operations:

Agent Development:

Key Features

Language Agnostic — Write agents in Go, Python, or any language with gRPC/HTTP support. The platform speaks protocols, not libraries.

Agent Discovery — Capability-based registry. Agents declare what they can do; the platform routes work to them.

Durable Streaming — LLM responses persist with sequence numbers. Clients reconnect and resume from where they left off.

Stateless Agents — Platform manages sessions, state, and memory. Agents connect, process, respond.

When Athyr Fits

Next Steps

Ready to get started? Head to the Installation guide or jump straight into the Quick Start tutorial.