Skip to content

Run your fleet of AI agents on Kubernetes

Each agent gets its own sandboxed pod: disk, tools, repos, and memory that survive restarts. Manage the whole fleet from a web console, an API, or straight from Telegram and Discord. Powered by the Claude or ChatGPT subscription you already pay for.
# Install from the published Helm chart. Nothing to pin, no fork.
helm install kyber oci://ghcr.io/matty-v/charts/kyber \
  --namespace kyber-system --wait
# 15 minutes from an empty cluster to a live fleet console
The Kyber fleet console: dashboard with agent status, per-agent context pressure, and a live terminal peek
What you can do

A real machine for every agent, not a chat window

Kyber treats agents as infrastructure: each one gets a durable, isolated environment it fully controls, and you get one console to run them all.

Long-lived, sandboxed agents

Each agent owns a whole pod with a persistent disk. Installed packages, cloned repos, credentials, and memory all survive restarts, upgrades, and preemption.

Any Kubernetes cluster

A Windows laptop, a spare box, a Mac, or any cloud. On GCP, Kyber provisions the VMs for you, preemptible instances included.

Chat from your phone

Two-way Telegram and Discord conversations with any agent in the fleet. Assign work, approve decisions, and get results without opening a terminal.

Own the whole environment

Choose each agent's machine, VM type, disk, CPU, and memory. Reboot it, stop it, or open a shell into the running agent straight from the console.

Schedules and handoffs

Cron inside each agent survives restarts. Agents send each other signed messages, so one agent's output becomes another's next prompt, and can wake it from sleep.

Your subscription, not per-token

Claude Code agents sign in with your Claude subscription, Codex agents with your ChatGPT subscription. API keys work too, but you do not need them.

How it works

From empty cluster to working fleet

Install the chart

One Helm install from the published OCI chart brings up the control plane, the console, and the API on any Kubernetes cluster, local k3d included.

Create your agents

Define each agent in the console: its runtime (Claude Code or Codex), its resources, its git-backed identity. Kyber schedules a sandboxed pod with a persistent disk.

Put them to work

Chat over Telegram or Discord, watch the fleet from the dashboard, put agents on cron schedules, and let them hand work to each other while you are away.

Use cases

What people run on Kyber

Kyber is the platform layer under the agents. Here is what fleets look like in practice.

Apache 2.0 licensed
v1.x, CRDs and API still evolving
Runs the maintainer's own fleets, every day
Built for clusters you trust your agents with

Give your agents somewhere to live

Fifteen minutes from an empty cluster, or a k3d one-liner, to a fleet console with a live agent. No cloud account required.