Skip to main content
Pulumi logo Pulumi logo
  1. Docs
  2. Infrastructure AI
  3. Pulumi Agents
  4. Work with tasks
  5. Use Neo in the CLI

Use Neo in the CLI

    Run pulumi neo to work with Neo from the terminal in your local project. You can also use your editor’s agent panel.

    Work in your local environment

    The agent can read and change local files and run installed CLIs. It uses your pulumi login account for Pulumi Cloud and credentials from your environment variables, CLI logins, and Kubernetes configuration for other systems. Sign in to the tools the task needs before starting a session. See Permissions and approvals for access and approval controls.

    Get started

    1. Authenticate to Pulumi Cloud with pulumi login.
    2. Run pulumi neo from a directory containing a Pulumi.yaml file.
    3. Describe the change you want or the problem to solve. Include the stack name and any relevant error output.

    See the pulumi neo command reference for the full list of flags and options.

    Controls

    Choose approval, permission, and plan settings for the session:

    • Approval modes (manual, balanced, auto) control when the session pauses for confirmation. manual is the same mode the console calls Review.
    • Permission modes (default, read-only) control whether the session can write to Pulumi Cloud. They do not restrict local file changes or the credentials used by local commands.
    • Plan mode lets the agent propose a plan before changing files (toggle with Shift+Tab before sending your first message).

    For connections configured in Pulumi Cloud, see Integrations.

    Agent handoff

    Install the delegation skill in your coding agent to transfer work to a Neo task. When you explicitly request a handoff, the skill summarizes the goal, repository state, and conversation, then passes that summary to pulumi neo.

    The handoff session can read your local working tree, including uncommitted changes. You do not need to commit or push them for the handoff.

    The skill returns a task URL and instructs the original coding agent to stop. Continue the work in that task. This is a one-way transfer; the original coding agent does not wait for a result and resume automatically.

      The infrastructure platform for humans and agents.