Manage Streaming Agents in the Confluent Cloud Console

You can create, configure, deploy, and monitor Streaming Agents directly in the Confluent Cloud Console without writing SQL. The Cloud Console provides a workflow for defining an agent’s instructions, model, and tools, and shows you deployments and configuration.

For the SQL-based workflow, see Create and Run Streaming Agents.

Prerequisites

You need the following before you can manage Streaming Agents in the Cloud Console:

A Confluent Cloud account.
A Apache Flink® compute pool in the target environment.
The FlinkDeveloper role or equivalent permissions for compute pool access, connection creation, and agent operations.
Write access to a Apache Kafka® cluster and topics for input, output, and log tables.
An LLM provider account such as OpenAI, Anthropic, or Google Gemini, and an API key to access the account.

Navigate to Streaming Agents

You can access Streaming Agents from two places in the Cloud Console.

From a Flink SQL workspace

To view Streaming Agents from a workspace:

Sign in to Confluent Cloud.
In the navigation menu, click SQL workspaces.
Open a workspace. If you don’t have a workspace, create one.
In the left panel, expand a cluster node.
Expand Agents.
If you have existing agents, they appear under the Agents node. You can also create an agent from here. For details, see Create an agent.

Create an agent

You create a Streaming Agent in a side panel where you configure the required and optional settings.

To open the creation panel, use one of the following methods:

On the Streaming Agents page, click Create streaming agent.
In the left panel of an SQL workspace, click the + button next to Agents.

To configure the agent, fill in the following fields in the creation panel:

For Database, select the cluster for the agent.
For Agent name, enter a name for the agent.
Optionally, for Description, enter a description.
For Instructions, enter the agent’s role, goals, and behavioral constraints.
Select or create a model. For details, see Select or create a model.
Optionally, click Show advanced configurations to add tools or set runtime options. For details, see Add tools and Configure advanced options.
Click Create agent.
Alternatively, click Get SQL to generate the equivalent Flink SQL statement instead of creating the agent directly.

Select or create a model

Every agent requires a model for reasoning. You can select an existing model, or you can create one when you create the agent.

Select an existing model

From the Model dropdown, select a model.

Create a new model

If you don’t have a model registered, create one. To create a new model:

In the Model dropdown, select Create new model.
For Provider, select a provider, such as OpenAI, Anthropic, or Gemini.
For Integration, select an existing connection, or create a new connection by providing the endpoint URL and credentials.
For Model name, enter a name.
Optionally, add or modify model properties like temperature and model version.
Click Save model.

The new model is auto-selected in the Model dropdown when you return to the agent creation form.

For details on model properties, see CREATE MODEL.

Add tools

Tools let an agent access external systems, perform actions, or retrieve information. You can add user-defined function (UDF) tools and MCP server tools in the advanced configurations section of the agent creation panel.

To add a tool to the agent:

Click Show advanced configurations.
Click Add new tool.
In the Add agent tools section, select an existing tool from the Agent tools dropdown, or click Create a new agent tool to create one.

Create a UDF-based tool

To create a UDF-based tool:

Click Create a new agent tool.
Select User defined function.
Select an existing UDF from the dropdown, or select Create new function and upload a Java JAR or Python artifact.
Click Create new agent tool.

The new tool is auto-selected in the tools dropdown when you return to the agent creation form.

Create an MCP server tool

To create an MCP server tool:

Click Create a new agent tool.
Select Custom MCP server.
For Integration, select an existing MCP connection, or click Add connection to create a new one by providing the MCP server endpoint and credentials.
Click Create new agent tool.

For more information on tool types, see Call Tools. For connection details, see CREATE CONNECTION.

Configure advanced options

The advanced configurations section includes the following runtime options:

Option	Description
Handle exception	How the agent handles errors. FAIL stops the agent on error. CONTINUE allows it to proceed.
Max iterations	Maximum number of tool-calling iterations.
Request timeout	Maximum time in seconds for each request.
Max consecutive failures	Number of consecutive failures allowed before the agent stops.
Tokens management strategy	How the agent manages conversation context when it grows too large. NONE performs no management. SUMMARIZE generates a summary. TRIM removes the oldest messages.

For the full configuration reference, see Agent Runtime Guide.

View agent details

To view the details for an agent, click the agent name on the Streaming Agents page. The details page has three tabs: Overview, Deployments, and Configurations.

Overview

The Overview tab shows the following information:

Catalog: The catalog where the agent is registered.
Database: The database where the agent is registered.
Model: The model that powers the agent.
Recent deployments: Active deployments or an empty state with a Run agent button.
Prompt: A summary of the agent’s instructions with a View full prompt link that opens the complete prompt in a window.
Tools: A list of tools attached to the agent. If no tools are attached, an Edit agent button links to the edit flow.

Deployments

The Deployments tab shows a table of all deployments for the agent with the following columns:

Target: The output table for the deployment.
Status: The deployment state.
CFU: Compute resource usage.
Type: The deployment type.
State size: The size of the deployment state.
Created: When the deployment was created.
Account: The account that created the deployment.

Use the search bar and status or type filters to find specific deployments.

Configurations

The Configurations tab shows a read-only view of the agent’s runtime configuration values, including handle exception, max consecutive failures, and tokens management strategy.

Edit an agent

You can modify an agent’s description, instructions, model, tools, and advanced configuration through the Cloud Console. To edit an agent:

On the agent details page, click Edit.
In the Edit streaming agent side panel, modify the fields as needed.
Optionally, click Show advanced configurations to update tools and runtime options.
Click Save change.

Important

Editing an agent does not affect existing deployments. Changes take effect when you recreate the deployment statement.

Deploy an agent

After you create an agent, you deploy it to process streaming data. To deploy an agent:

On the agent details page, click Run.
In the Run streaming agent panel, review the pre-filled SQL steps for creating the input table, output table, and agent system table.
Click Done to open a Flink SQL workspace with the deployment query.

The deployment creates a Flink job that reads events from the input table, processes them through the agent, and writes results to the output table. You can track the deployment status on the Deployments tab of the agent details page.

Monitor an agent

The Deployments tab on the agent details page provides real-time status for running agent deployments, including resource usage and state information.

Delete an agent

To delete an agent:

On the agent details page, click More.
Click Delete.
In the Confirm deletion window, enter the agent name to confirm.
Click Confirm.