Agentic Data Plane

Agentic Data Plane Learning Path

Use this page to learn Redpanda Agentic Data Plane in an order that builds on itself, whether you set up the platform, build agents, or govern them. The rest of the docs group pages by task. Here, the same pages run from how the platform works, to your first running agent, to governing agents and connecting them to your production systems.

Each of the four levels lists what you can do when you finish it, so you can start at the level that fits your role and skip what you already know.

After reading this page, you will be able to:

  • Identify where to start with Agentic Data Plane for your role

  • Recognize what you can do with Agentic Data Plane after each level, from running your first agent to governing agents in production

  • List the controls Agentic Data Plane gives you to govern agents and their spend

Choose where to start

Your role Start here

Administrator or platform engineer

Agent builder

Security or compliance lead

Evaluator or pilot lead

The four levels

Level When you finish it, you can Before you start

Level 1: Understand the platform

  • Name the components and who runs each

  • Explain how the platform gives an agent an identity, sits between the agent and every model and tool it calls, and records what it did

  • Tell a managed agent or MCP server from a self-managed one

Nothing. Start here if you have never used the product.

Level 2: Build and run your first agent

  • Configure a LLM provider

  • Create and test an MCP server

  • Build an agent and test it in the Playground

  • Read the agent’s transcript

  • Level 1

  • Access to an Agentic Data Plane environment, which an administrator can create if you don’t have one

  • For administrators, credentials for an LLM provider, such as the Anthropic API key the administrator quickstart uses

  • If your environment has a Private label, read Private Environments first

Level 3: Govern agents

  • Decide who can invoke which agent or tool

  • Tag resources and manage secrets

  • Block or mask content with guardrails

  • Contain a misbehaving agent

  • Cap and analyze spend

  • Shape what tools return

  • Trace a slow or failed run

  • Read the audit log

Level 2 and a running agent to govern.

Level 4: Integrate with production

  • Connect your own agents, applications, and AI clients to the platform

  • Manage the platform’s resources as code

Level 2, plus the parts of Level 3 your organization enforces.

Level 1: Understand the platform

Level 2: Build and run your first agent

Start with the quickstart for your role, then read about the pieces it used.

Topic Read

Quickstarts

Agents

MCP servers

Level 4: Integrate with production

Take the tracks your project needs.

Track Read When you finish it, you can

Your own agents and apps

  • Write a system prompt that holds up under real traffic

  • Choose an agent architecture

  • Point your own application or coding assistant at AI Gateway

  • Register an agent you run yourself so it shows up in transcripts

Agents in your systems

  • Invoke an agent over the A2A protocol from an application, or from an agent hosted outside Agentic Data Plane

  • Start an agent from Microsoft Teams or on a schedule

Authentication and external clients

  • Let each user reach an upstream system with their own credentials

  • Register an OAuth provider

  • Bring your own MCP server under the gateway

  • Connect chat apps and IDEs to your servers

  • Add your servers to your Claude organization as custom connectors

Resources as code

  • Manage providers, servers, agents, and policies from the CLI

  • Keep them in version control

Keep up with changes

The product changes often. Read the Release Notes for each version, and reread the pages for the product areas you use. A Preview label marks a feature that isn’t generally available (GA) yet, so its behavior can change.