YOUR AGENTS. OFF YOUR LAPTOP.

Close your laptop.
Keep your agents working.

A home for your coding agents.
A workspace you control.

Run Claude Code, Codex, Cursor, OpenClaw or OpenCode in isolated workspaces on your own Kubernetes cluster. Give each session its repos and tools, schedule the next task, and supervise from anywhere.

Start self-hosting ↗ Star on GitHub

AGPL-3.0 core · Bring your own infrastructure and agent login

Explore the workflow · Read the field notes →

YOUR CLUSTER / AGENT WORKSPACE SESSION
TONIGHT'S TASK

Fix the flaky test.

autonomous
↳
You set the task

Selected repo. Project toolchain. Scoped credentials.

01
✓
The agent works in its own pod

Inspect the code, make changes, run tests.

02
↗
You review the result

Read the session. Check the diff. Decide what ships.

03
# Your laptop can sleep.
agent workspace → running on your cluster
Illustrative workflow · outcomes depend on the task and agent
BUILT FOR YOUR STACK

Claude Code / Codex / Cursor / OpenClaw / OpenCode

KUBERNETES + YOUR TOOLCHAIN

01 / A WORKSPACE THAT STAYS

Less tied to your laptop.
More control over the work.

For developers and platform teams ready to run agent workloads on their own infrastructure.

01 — PERSISTENCE

Keep the task running.

Interactive, autonomous or scheduled sessions. Start a task, close your browser, and return to the workspace from another device.

Pair interactivelyRun unattendedSchedule a session
02 — ISOLATION

A pod for each session.

Separate workspaces, selected credentials and network policies. Give a session the access its task needs.

03 — YOUR TOOLCHAIN

Bring the real environment.

Connect your Git repo, choose a container image and attach MCP servers for the tools you already use.

04 — HUMAN SUPERVISION

Stay close to the work.

Follow the terminal from your phone. Open the integrated browser when the task needs a visible browser and human collaboration.

05 — COORDINATION

Give each agent a focused job.

Separate research, implementation and review into sessions. Use agent messages and the API to build your own handoffs and acceptance checks.

Explore the building blocks ↗

02 / FROM TASK TO REVIEW

Your next task.
Its own workspace.

Start with one useful task. Add sessions and integrations when the workflow calls for them.

  1. 01

    Connect the essentials

    Add repo access and your agent subscription login or API key. Choose only the tools this task needs.

  2. 02

    Set the task and mode

    Choose the agent, image and repo. Pair interactively, hand off an autonomous task or schedule recurring work.

  3. 03

    Return and review

    Inspect the session and its output. Continue a finished autonomous run interactively when you need to follow up.

BUILD YOUR OWN FLOW

Research, ticket handling and multi-agent review are possible with the right tools and integration. You define the workflow and what needs human approval.

03 / MAKE IT YOURS

Your cluster.
Your starting point.

Already running Kubernetes? Deploy with Helm. Getting started locally? The quickstart provisions a single-node cluster.

EXISTING KUBERNETES CLUSTER

Deploy with Helm

helm repo add agenthub \
  https://open-agenthub.github.io/open-agenthub
helm install agenthub agenthub/open-agenthub \
  -n agenthub --create-namespace \
  --set postgres.password="$(openssl rand -hex 16)"

Keep the generated database password for upgrades. Follow the guide to configure storage, access and authentication.

Read the deployment guide ↗
NO CLUSTER YET

Use the local quickstart

Linux uses k3s. macOS and Windows use k3d with Docker Desktop. The installer includes persistent Postgres and object storage.

Linux / macOS
curl -fsSL https://open-agenthub.github.io/install.sh | sh
Windows PowerShell
iwr -useb https://open-agenthub.github.io/install.ps1 | iex

Recommended starting resources: 4 vCPU · 6 GB RAM. Requirements grow with your workloads.

From version 1.5.0, installs without OIDC require a shared admin API key. Use HTTPS when exposing your instance; configure OIDC for individual identities. Read the setup guide and upgrade instructions. Infrastructure and model-provider usage are separate costs.

04 / OPEN CORE. CLEAR COSTS.

Self-host the core.
Start with a real task.

The Community core is AGPL-3.0. You operate the infrastructure and bring your agent subscription or API key.

An Enterprise edition adds team features such as session sharing, organisation controls and Slack integration.

COMMUNITY

No core licence fee.

AGPL-3.0 · self-hosted

  • Interactive, autonomous and scheduled sessions
  • Your repos, container images and MCP tools
  • Integrated browser and remote supervision
  • Telegram and Signal integrations
Start with Community ↗

Infrastructure, provider charges and configured usage limits still apply.

05 / BEFORE YOU DEPLOY

Good questions.
Clear answers.

Do I need to know Kubernetes?

Open AgentHub runs on Kubernetes. The quickstart can create a local cluster for you, but you still operate the deployment. If you already manage a cluster, use the Helm chart.

Does self-hosting keep all model data local?

No. The agent workspace runs on your infrastructure, but prompts and other model inputs can still go to your chosen provider. Check the provider's data policies for your use case.

What does session isolation cover?

Each session runs in its own pod with selected credentials and network policies. Containers share the host kernel unless you configure an additional runtime boundary. Isolation does not replace careful credential scoping.

Can I build a ticket-to-agent workflow?

Yes, using the API, MCP tools and session messaging. You build the integration, handoffs and review gates. Open AgentHub is not a preconfigured service-desk automation product.

Can I try a hosted demo?

There is currently no hosted trial. Start with the local quickstart or deploy to your cluster. The getting-started guide walks through setup.

READY FOR THE NEXT TASK?

Give your agent
a place to work.

Read the quickstart ↗Explore the source ↗