Skip to main content
Welcome to the Flokoa documentation. Flokoa lets you deploy, configure, and scale AI agents on Kubernetes using familiar declarative YAML — the same way you manage any other workload in your cluster. Whether you’re running a single prototype or a fleet of production agents, Flokoa gives you the operational primitives to do it reliably.

Introduction

Learn what Flokoa is, how it works, and the six Custom Resource Definitions that power your AI agents.

Quickstart

Install the operator, connect an LLM provider, and deploy your first agent in under five minutes.

Agent Resource

Explore the full Agent CRD spec — runtime modes, model references, tool bindings, and lifecycle options.

SDK Overview

Use the Flokoa Python SDK and CLI to build, run, and interact with agents programmatically.

​
Get up and running

Follow these four steps to go from a fresh Kubernetes cluster to a running AI agent.
1

Install the operator

Apply the single-file install manifest to your cluster. The Flokoa operator is deployed into the flokoa-system namespace and immediately begins watching for Flokoa Custom Resources.
2

Configure an LLM provider

Create a Kubernetes Secret with your API key, then declare a ModelProvider resource pointing to OpenAI, Anthropic, Google Gemini, or AWS Bedrock. A Model resource wires a specific model name and parameters to that provider.
3

Deploy your first agent

Apply an Agent resource that references your Model and specifies a container image. The operator creates a Kubernetes Deployment and Service automatically — no Helm chart or manual pod spec required.
4

Call it via A2A

Once the agent’s status shows Running, reach it through its cluster-internal Service endpoint using the Agent-to-Agent (A2A) protocol or any HTTP client.

​
Explore the resource types

Flokoa exposes six Custom Resource Definitions under the agent.flokoa.ai/v1alpha1 API group. Each resource plays a distinct role in the agent deployment pipeline.

Agent

The primary resource. Declares the agent runtime, model reference, tool bindings, and scaling configuration.

ModelProvider

Stores LLM provider connection config and API credentials, shared across any number of Model resources.

Model

Selects a specific model name and tuning parameters, referencing a ModelProvider for the connection.

AgentTool

Gives agents access to external APIs and internal Kubernetes services via OpenAPI specifications.

AgentWorkflow

Composes multiple agents into multi-step workflows, compiled to Argo WorkflowTemplates automatically.