Distributed Claude AI agents on Kubernetes — first-class cluster resources via CRDs and operators.

komputer.ai is a Kubernetes-native platform for running persistent Claude AI agents as first-class cluster resources. No external database — KomputerAgent CRs live in etcd, the operator reconciles them into Pods, and the API streams every event back over WebSocket.

Quick start

helm install komputer-ai oci://ghcr.io/komputer-ai/charts/komputer-ai \
  --set anthropicKeySecret.name=anthropic-api-key \
  --namespace komputer-ai --create-namespace

That deploys the operator, API, UI, Redis, and a default agent template. From there: port-forward, open the dashboard, create your first agent. Full walkthrough in Installation.

Get started

Installation

Helm install on any Kubernetes cluster — five minutes from zero to running agent.

Architecture

Components, data flow, protocols, tech stack, and the design choices behind the platform.

Integration overview

Drive komputer.ai from your own services via REST + WebSocket.

Core concepts

Agents

The central entity — lifecycle modes, roles, overrides, concurrency control.

Squads

Multi-agent shared workspaces via co-located Pods. Manager + worker collaboration.

Templates

Reusable Pod specs for agents — image, resources, storage, service account.

Connectors

MCP server connections per agent or template; includes the supported-services compatibility matrix.

Power features

Skills

Reusable capabilities attached to agents as Kubernetes CRs — written into the workspace at start.

Memories

Persistent knowledge injected into the system prompt. Per-agent or per-namespace.

Schedules

Cron-based recurring tasks with timezone support and auto-cleanup.

Cost tracking

Real-time per-task cost, context window monitoring, per-agent breakdown with task-level drill-down.

Build with it

REST API

Create agents, send tasks, list status, fetch results — full control plane in HTTP.

WebSocket events

Stream tokens, tool calls, costs, and lifecycle transitions in real time as agents work.

Custom agent images

Bake your own packages, CLIs, or runtimes into the agent image.

Operate it

Monitoring

Prometheus metrics, ServiceMonitors, and a Grafana dashboard ship with the chart.

Logging

Structured JSON or human-readable logs across all components.

Why komputer.ai

Kubernetes-native — agents are CRDs reconciled by an operator. Pods, PVCs, lifecycle, GitOps — all standard primitives.
Stateless API — no database. CR .status is the source of truth; Redis is just transport.
Real-time streaming — REST for control, WebSocket for live token-by-token output, tool calls, and costs.
Squads & sub-agents — co-located multi-agent pods. Manager agents create and orchestrate workers via MCP tools.
Built-in observability — structured logs, Prometheus metrics, and a Grafana dashboard ship with the chart.
MCP connectors — plug in MCP servers per agent or template; secrets and lifecycle managed by the operator.

Welcome