Kelos CLI
A CLI and Kubernetes operator that runs Claude Code, Codex, Gemini, OpenCode and Cursor as tasks and sessions inside a cluster
- 354 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- Agents get a repository, credentials and the right to open pull requests on events in the cluster without a human
- Agent keys and OAuth tokens sit in a cluster Secret, and the skill creates and replaces them
- The operator installs CRDs and a controller cluster-wide and creates Pods, Jobs and StatefulSets
Install
- Works in
Manual install
curl -fsSL https://raw.githubusercontent.com/kelos-dev/kelos/main/hack/install.sh | bashInstalls the kelos CLI. Needs a Kubernetes 1.28 or newer cluster with cert-manager.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Agents get a repository, credentials and the right to open pull requests on events in the cluster without a human
- Agent keys and OAuth tokens sit in a cluster Secret, and the skill creates and replaces them
- The operator installs CRDs and a controller cluster-wide and creates Pods, Jobs and StatefulSets
This is third-party code. Review the repository files before installing.
What it does
Kelos turns coding agents into Kubernetes workloads: a controller translates its own resources into Pods, Jobs and StatefulSets. It supports Claude Code, OpenAI Codex, Gemini, OpenCode, Cursor and custom agent images. A Task runs one job, a TaskPipeline sequences stages with matrix fan-out, a Session keeps a live conversation with an agent, a Workspace gives the agent a Git repository, and an AgentConfig holds shared instructions, skills, plugins and MCP servers. Runs can be tied to events: a TaskSpawner creates tasks from GitHub, Jira or Linear events, a schedule or a webhook, a SessionSpawner creates sessions from GitHub webhooks, and a WorkerPool keeps ready workers. Kelos records branches, commits, pull requests and token usage in the task status. The kelos command creates workspaces, runs tasks and reads their logs, and an optional web console shows session chats in a browser. The kelos skill from the repository (npx skills add kelos-dev/kelos) teaches an agent to install Kelos, write manifests and troubleshoot failures, so you do not have to write YAML by hand.
Who it is for. Platform and DevOps teams with a Kubernetes 1.28 cluster and cert-manager who need to run agents away from developer laptops: on events, on a schedule and in parallel across repositories.
Good fit when
- Agents run in isolated cluster Pods and developer laptops stay out of it
- An issue labeled agent-ready turns into a task by itself: reproduce the bug, add a test and open a pull request without merging
- One change runs in parallel across ten repositories through a TaskPipeline with a matrix
- A long session with an agent stays alive and you reconnect to it from a terminal or a browser
- Agent spend and limits are controlled with ordinary Kubernetes tools
Not a fit when
- There is no Kubernetes 1.28 or newer cluster with cert-manager: without them the operator will not install
- You need a single agent on your own laptop: Kelos is too heavy for that
- Agent keys and OAuth tokens cannot be kept in a cluster: Kelos hands them to tasks through a Secret
- You need a stable resource schema: the API lives in the kelos.dev group at version v1alpha2 and fields may change
Example request
Using the /kelos skill, create a Codex Task in my-workspace with the prompt "Add a hello world program in Python", run it and show me the logsLimitations
Installing needs a Kubernetes 1.28 or newer cluster and cert-manager. Resources live in the kelos.dev group at API version v1alpha2, so fields may change. Agent keys and tokens are kept in a cluster Secret, and the author advises scoped credentials, protected branches and workload limits. The kelos skill is meant for working with Kelos resources and commands and does not take on ordinary code edits in a repository. The web console is enabled separately and requires a shared token.
How to disable. The README has no uninstall command. Kelos resources are deleted with ordinary kubectl delete, and the operator is removed the way it was installed: through Helm or the kelos install command. The skill is deleted from your agent's skills folder.
FAQ
What does Kelos need to install?
A Kubernetes 1.28 or newer cluster with cert-manager installed, and the kelos CLI. The CLI installs by script, through Homebrew (kelos-dev/tap) or go install, and kelos install puts the operator in the cluster.
Which agents does Kelos run?
Claude Code, OpenAI Codex, Gemini, OpenCode and Cursor. Custom agent images are supported through an interface described in the repository.
How do I run agents on events?
Through a TaskSpawner: it creates tasks from GitHub, Jira or Linear events, a schedule or a webhook. In the README example tasks are created for issues labeled agent-ready, with concurrency 3 and a 1-hour runtime.
How do I open the web console?
The console is optional: you enable it in the Helm chart and set a shared token. The interface runs inside the cluster, shows Kelos resources and lets you create sessions and chat in them.
Can I avoid hand-written YAML?
Yes. The kelos skill installed with npx skills add kelos-dev/kelos writes and applies manifests from a description of the outcome. The same resources are also created with kelos create and kelos run.
Discussion
No questions or reviews yet.
Related
CLI toolsKubeShark skill: the agent finds the failure mode first, then writes manifests, Helm and Kustomize to official practice with a rollback plan
OpenCode agents on Kubernetes: tasks and shared settings live in manifests, and kubeoc attaches to a live agent
A self-hosted control center for coding agents: Claude Code, Codex, Copilot, OpenCode and Pi from your phone, desktop and terminal
A Kubernetes panel with an MCP server: Claude Code reads pod logs, edits resources and rolls back releases in connected clusters.
31 DevOps skills for Claude Code and Codex: config generators, validators and a Kubernetes debugger for Terraform, Helm, CI and Docker
An AI assistant for Kubernetes that turns plain-language requests into kubectl operations and doubles as an MCP server
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Kelos (this card) | no | 354 | ||
| CLI Agent Orchestrator | no | 1,515 | coordinates agents in tmux sessions on one machine, while Kelos runs them as Pods in a Kubernetes cluster | |
| Ralph Orchestrator | no | not required | 3,194 | loops an agent until done on one machine, while Kelos works through cluster resources and events |
README in short
The page opens with the idea of running agents as Kubernetes workloads and a list of reasons. A table of resources follows, then a quick start with the skill and by hand through the CLI, console features and a list of scenarios: developer-guided sessions, event-driven workers, pipelines, repository fleets and the project's self-development loops.