kynomesh

Kynomesh

Welcome to Kynomesh! A Kubernetes-native Agent Mesh platform for orchestrating distributed multi-agent systems. Kynomesh lets you declare a group of cooperating AI agents as a single Kubernetes resource and takes care of placement, peer discovery, and agent-to-agent (A2A) traffic - so you can focus on the agent logic instead of the wiring.

Each agent runs in its own dedicated set of pods, accompanied by an injected broker sidecar that manages A2A communication and transport. Kynomesh provides built-in autoscaling capabilities, automatically scaling agent workloads based on traffic patterns and system load. Agents are language-agnostic and can be implemented using any programming language.

Use Cases

• Multi-agent LLM applications: a supervisor coordinating worker agents, or a swarm of specialized agents that hand off tasks to one another.
• Sequential agent pipelines: each stage produces input for the next.
• Heterogeneous agent fleets: combine agents written in different languages behind a single Kubernetes API.
• Hybrid topologies: mix in-cluster managed agents with external agents reachable at a URL.

Key Features

• Kubernetes-native: AgentSet is the primary custom resource in Kynomesh. If you're familiar with kubectl, you'll feel right at home managing and operating agent workloads.
• Pattern-based peer discovery: Define a communication pattern (such as Supervisor, Handoff, or Sequential) along with an entry agent, and Kynomesh automatically derives and maintains peer relationships. No custom service-discovery logic is required in your agents.
• Language agnostic: Build agents in any programming language, including Go, Python, and more.
• Built-in autoscaling: Automatically scales agent workloads in response to traffic demand and system load, ensuring efficient resource utilization and high availability.
• Zero-downtime rolling updates: Kynomesh orchestrates rolling deployments and graceful pod termination, enabling seamless upgrades with minimal disruption to ongoing traffic.
• Rate limiting: Cap concurrent in-flight A2A requests, protecting agents (and their downstream dependencies) from overload.
• Hybrid topologies with external agents: Combine in-cluster managed agents with external, HTTP-reachable agents behind a single Kubernetes API and unified peer discovery.
• Automatic drift reload: Kynomesh detects when a peer agent's AgentCard changes and automatically restarts only the pods still holding a stale cached copy, so dependent agents pick up the change without a manual restart.

Resources

• Quick Start
• AgentSet
• Tutorials
• Code Examples
• Autoscaling
• Rate Limiting
• Drift Reload
• External Agents
• SDKs [Python | Go]
• Development
• Static code analysis
• Contributing

License

Apache License Version 2.0, see LICENSE.