Welcome to IBM Paver

Learn about IBM® Paver, an AI-native infrastructure operator embedded across IBM Infrastructure platforms and services. It is purpose-built for the way teams work on shared systems where operational responsibility, approvals, and oversight matter. Rather than just answering questions, Paver is designed to take action on your infrastructure.

View as markdown

On this page

Describe the outcome that you need. Paver maps that request to your current environment, builds a plan, evaluates risk, and executes through your existing infrastructure tools and workflows with human approval required before any consequential change takes effect.

What Paver does#

Paver runs a governed operating loop that takes your request from interpretation through planning, risk assessment, approval, and execution:

Interprets the request and environment by mapping the outcome that you need to the current infrastructure state.
Plans and explains by building a plan, describes what changes before they happen, and surfaces dependencies before acting.
Evaluates risk with Paver Guard by identifying consequential actions and requires human approval when production state might change.
Executes through your existing tools by working through the CLI, APIs, scripts, Terraform, and the platform tools your team already uses.
Retains operational context by storing relevant context so future operations build on what Paver already established about your environment.

All operations are bounded by your account permissions.

Key capabilities#

Skills

Add expertise to Paver for your team's workflows and specialized knowledge to get more accurate guidance, actions, and recommendations for your requests.

Shell

Use an interactive terminal in the Paver sandbox that you can use to run commands directly, alongside the agent, without leaving your session.

Keyring

Securely store a secret once, export it, and Paver makes it available in your session, without requiring you to paste credentials into the chat.

Memory

Take advantage of persistent context that Paver builds automatically across conversations so Paver knows how you like to work without the need to repeat yourself.

Paver Guard

Set policy-based approval requirements that control when and how Paver can change your infrastructure. Paver Guard evaluates proposed operations for risk and routes consequential actions for human approval before execution.

Safety mode

Control whether Paver can make changes in your environment in a session by switching between Safe mode, which prevents changes, and Edit mode, which enables approval workflows.

How Paver helps your team#

Paver is designed for infrastructure teams that need more than guidance. They need safe, reliable execution that accelerates day-to-day operations:

Triages and resolves incidents faster: Correlate logs, metrics, configuration, and recent state changes to investigate issues and speed recovery.
Reduces manual operational overhead: Automate multi-step routine tasks and prevent configuration drift across team workflows.
Scales operational consistency: Standardize how common infrastructure tasks are executed across team members of varying experience levels.
Continuously optimizes environments: Surface actionable opportunities to improve security posture, performance, resiliency, and resource efficiency.

It's built for environments where reliability, governance, and accountability are nonnegotiable. It runs within your existing security policies and is bounded by your authorized access, so it can't exceed your permissions.

Teams can adopt Paver incrementally. Start by using Paver for exploration, planning, and recommendations to verify outputs before enabling execution. When you're ready, configure Paver Guard policies to enforce human approval for consequential changes in production.

Getting started with Paver#

Paver is embedded across supported IBM Infrastructure offerings.

Offering	Description
IBM Cloud®	An agentic AI operator built into IBM Cloud that helps you deploy, operate, and troubleshoot your IBM Cloud account and infrastructure resources through natural language. Paver is available through the console and IBM Paver CLI. For more information, see Getting started with IBM Paver.
IBM Power Autonomous Operations	An AI-enabled operations platform that brings agentic capabilities to the management of mission-critical Power environments. The offering augments IT teams through a human-in-the-loop model, combining operator expertise with intelligent, policy-driven agents that assist in planning, execution, and optimization.

Note

Paver capabilities might vary across offerings. Refer to the documentation for the specific offering where you're using Paver for more information.

Trying out the IBM Paver CLI with IBM Cloud#

You can try out Paver at no cost by using the IBM Paver CLI with IBM Cloud. You need an IBM Cloud account to log in after installing the IBM Paver CLI. If you don't have one, go ahead and create an IBM Cloud account.

To install the IBM Paver CLI on macOS or Linux, run the following command:

curl -fsSL https://paver.cloud.ibm.com/cli/install | bash

After you install the CLI, run paver login to log in with your IBM Cloud credentials.

This command is for macOS and Linux only. For Windows steps and more information about installing the CLI and logging in, see Installing the IBM Paver CLI.

Starting your first chat#

Depending on where you're using Paver, you can start a chat from the web UI or CLI. From the web UI, choose one of the starter prompts or ask Paver how it can help you. From the CLI, use the paver chat command to get started.

If you have an IBM Cloud account, you can try your first chat in the web UI by going to Paver in the IBM Cloud console.

You can ask Paver about its features, capabilities, and common use cases, or start planning a specific infrastructure task. After your session starts, type a question or request in natural language. Here are some tips for chatting with Paver:

Give me a quick tour of Paver capabilities
What are the best practices for getting started with IBM infrastructure?
Can you help me identify resources that are rarely used?
Set up monitoring and alerting for my infrastructure
Review my infrastructure costs and suggest optimizations

Note

Your prompts, context, and operational data are never used to train foundation models.