Language Operator

Deploy natural-language workloads to Kubernetes.

Language Operator extends Kubernetes with purpose-built CRDs for AI agents. The workload is natural language -- you describe what you want done, declare the models and tools the agent can use, and apply it like any other manifest. The operator runs it on a runtime you already know: Claude Code, OpenClaw, OpenCode, Kilo, Qwen Code, Cursor, Goose or DeepAgents.

Each agent is an ordinary Argo Workflow — always-on, or run on a schedule — scheduled by the control plane, observable with the tools you already run, and managed through the same GitOps workflows as the rest of your cluster. No new framework to adopt and no code generation: you write the intent, and the operator wires up models, tools, config, networking, and storage.

Language Operator has purpose-built CRDs for agentic workloads:

Declare what you want in a manifest; the operator reconciles it into a WorkflowTemplate and the Workflow or CronWorkflow that runs it, adds a Service and NetworkPolicy, injects the model and tool endpoints, mounts the instruction and persona config, and keeps it that way as your cluster changes.

A fully-autonomous deepagents task for data analysis:

apiVersion: langop.io/v1alpha1
kind: LanguageAgent
metadata:
  name: data-analyst
spec:
  runtime: deepagents
  instructions: |
    Analyze the orders database and email a summary of monthly revenue
    trends, the top ten customers, and any anomalies to analytics@example.com.

    Work read-only and keep it cheap: inspect the schema and indexes first,
    check each query with EXPLAIN, and avoid full table scans on orders.
  models:
    - name: claude-sonnet
  tools:
    - name: orders-postgres-db
    - name: email

A Claude Code agent with the repository cloned into its workspace and Context7 for up-to-date library docs:

apiVersion: langop.io/v1alpha1
kind: LanguageAgent
metadata:
  name: workstation
spec:
  runtime: claude-code
  repository:
    url: https://github.com/your-org/your-service
    secretRef:
      name: github-credentials
  tools:
    - name: context7

Same as above, but with default instructions:

apiVersion: langop.io/v1alpha1
kind: LanguageAgent
metadata:
  name: maintainer
spec:
  runtime: claude-code
  repository:
    url: https://github.com/your-org/your-service
    secretRef:
      name: github-credentials
  instructions: |
    Triage open issues labeled "good first issue", implement the fix on a
    branch, run the tests, and open a pull request.
  tools:
    - name: context7

examples/development-team deploys a self-managing maintainer agent: on each scheduled run it picks the next open issue, implements it, runs the tests, and opens and merges a pull request — one issue per run.

Follow the installation guide and quick start to get started.

See the development setup guide for full instructions.

Alpha — APIs may change between releases.

MIT