Skip to main content
Version: 1.42

Agentic Workflows

AI agents are most effective when they can deploy code, run tests, and get feedback from real environments, not just read and write files locally. Okteto gives agents isolated, live environments where they can validate their work using the same CLI and okteto.yaml manifest that human developers use. No special infrastructure or per-repo configuration needed.

This section covers how to connect AI agents (such as Claude Code) to your Okteto environments, whether you're working alongside an agent in your IDE or letting it handle a task end-to-end.

note

This is different from Okteto AI, which is a standalone product for launching and managing agents from the Okteto Dashboard. Agentic workflows are about bringing your own agent tooling to work with Okteto environments directly.

Why Okteto for agents​

Each agent gets its own Okteto namespace with dedicated compute, storage, and network, so agents can't interfere with each other or with production.

Agents read okteto.yaml to discover which services exist, how they're built, how they're deployed, and what tests are available. No hardcoded service lists or extra config files to maintain. The same okteto.yaml that developers use is the only thing the agent needs.

The CLI commands are identical too: okteto deploy, okteto build, okteto test, okteto logs. If it works for a developer, it works for an agent.

Agents can deploy changes, hit live endpoints, read logs, and run tests against real infrastructure. And every PR can get its own running environment that stakeholders can access without any local setup.

Two operating modes​

Collaborative mode​

A developer works alongside an agent in their IDE or terminal. The developer manages the dev environment with okteto up, and the agent runs commands inside it using okteto exec. This is the way to go when you're pair-programming with an agent, doing interactive debugging, or working on something where you want to stay in the loop.

See Collaborative Workflows for details.

Autonomous mode​

An agent handles the full lifecycle, from reading a ticket to opening a pull request, without human intervention. It uses okteto deploy, okteto build, and okteto test to manage the environment directly. This works well for ticket-to-PR automation, CI-triggered pipelines, or batch tasks like writing tests across a codebase.

See Autonomous Workflows for details.

The okteto up rule​

okteto up starts an interactive development session with file sync and a live terminal. It requires human interaction and will hang indefinitely if an agent runs it.

Agents must never run okteto up.

  • In collaborative mode, the human runs okteto up. The agent uses okteto exec to run commands in the active dev container.
  • In autonomous mode, the agent uses okteto deploy + okteto build to manage services directly. okteto up is not part of the workflow.

Getting started​

Install the Claude Code Plugin for Okteto:

/plugin marketplace add okteto/okteto-claude-plugins
/plugin install okteto

This gives your agent built-in knowledge of Okteto CLI commands, workflow patterns, and auto-discovery from okteto.yaml. Then choose your workflow: