Build AI with Confluent Intelligence in Confluent Cloud

Confluent Intelligence is a suite of capabilities for building agentic AI workflows with Flink SQL in Confluent Cloud. You can create streaming agents, serve real-time context to AI agents, run built-in ML functions, and connect to remote AI models.

Streaming Agents

Streaming Agents bridge the gap between enterprise data and AI capabilities. You can create AI agents that run continuously in Flink SQL, connecting your streaming data to AI models. With Streaming Agents, you can:

Access real-time data for AI decision-making.
Integrate with any tool, model, and data system.
Enable agents to plan, decide, and act on live operational events.

For more information, see Streaming Agents.

Real-Time Context Engine

The Real-Time Context Engine delivers real-time data from your Apache Kafka® topics to external AI agents through MCP. When you enable the Real-Time Context Engine on a topic, Confluent Cloud materializes the topic data into a table optimized for fast lookups. AI agents query the data through MCP tools without interacting directly with Kafka.

For more information, see Real-Time Context Engine.

Built-in machine learning (ML) functions

Built-in ML functions with Confluent Cloud for Apache Flink

Built-in ML functions simplify complex data science tasks into Flink SQL statements. You can run forecasting, anomaly detection, sentiment analysis, and PII detection directly in Flink SQL, with no ML expertise or model building required.

The following functions are available:

AI_DETECT_ANOMALIES: Detect anomalies in your data using foundation models.
AI_FORECAST: Forecast future values in time-series data using foundation models.
AI_SENTIMENT: Analyze sentiment toward specific aspects in text.
ML_DETECT_ANOMALIES: Detect anomalies in your data.
ML_EVALUATE: Evaluate the performance of an AI/ML model.
ML_FORECAST: Forecast trends in your data.
ML_PREDICT: Run a remote AI/ML model for prediction, text generation, and classification.

ML preprocessing utility functions are also available:

ML_BUCKETIZE, ML_LABEL_ENCODER, ML_ONE_HOT_ENCODER: Encode and bucketize columns.
ML_MAX_ABS_SCALER, ML_MIN_MAX_SCALER, ML_ROBUST_SCALER, ML_STANDARD_SCALER, ML_NORMALIZER: Scale and normalize columns.
ML_CHARACTER_TEXT_SPLITTER, ML_FILE_FORMAT_TEXT_SPLITTER, ML_RECURSIVE_TEXT_SPLITTER: Split text into chunks.
ML_NGRAMS: Create n-grams.

For more information, see Built-in AI/ML Functions.

Supporting technologies

Confluent provides the following technologies for building AI/ML workflows.

Remote and managed model inference

Remote model inference with Confluent Cloud for Apache Flink

You can run inference with remote AI/ML models in Flink SQL. Connect to models hosted on OpenAI, AWS Bedrock, AWS Sagemaker, Google Cloud Vertex AI, and Azure AI Foundry. You can also run fully managed AI models in Confluent Cloud.

For more information, see Run a Remote AI Model and Run a Managed AI Model.

External tables and search

External tables and search with Confluent Cloud for Apache Flink

You can enrich data streams with non-Kafka data sources by using external tables. Join real-time data streams with data from relational databases, vector databases, and REST APIs to enable retrieval-augmented generation (RAG) and more accurate AI decision-making.

For more information, see Search External Tables.

Real-time embedding support

You can continuously turn unstructured enterprise data into vector embeddings to enable RAG and mitigate LLM hallucinations. Use any embedding model and any vector database across any cloud.

For more information, see Create Embeddings.

Secure connections

Reusable connection resources provide a secure way to integrate with external systems. You can connect to models, vector databases, and MCP servers using Flink SQL. Sensitive credentials are stored separately from connection metadata and are never exposed in catalog metadata, logs, or configuration files.

For more information, see Reuse Connections.

Flink SQL integration

Confluent Cloud for Apache Flink provides a Flink SQL interface for creating and managing model, agent, and tool resources. The SQL interface is available in Cloud Console and the Flink SQL shell.

You can use the following SQL statements to create resources:

The following functions are available for model inference and agentic AI workflows:

AI_TOOL_INVOKE: Invoke MCP tools and user-defined functions (UDFs).
AI_COMPLETE: Generate text completions.
AI_EMBEDDING: Create embeddings.
ML_PREDICT: Run an AI/ML model for prediction, text generation, and classification.

For more information, see Run an AI Model.

RBAC for model inference

The following table shows the model actions that are available for different RBAC roles.

Role	CREATE MODEL	Invoke model for prediction	List/Describe Models	DROP MODEL	Grant permissions on models
OrganizationAdmin	Yes	Yes	Yes	Yes	Yes
EnvironmentAdmin	Yes	Yes	Yes	Yes	Yes
CloudClusterAdmin	Yes [1]	Yes [1]	Yes [1]	Yes [1]	Yes [1]
ModelDeveloperManage	Yes	No	Yes	Yes	No
ModelDeveloperRead	No	Yes	Yes	No	No
ModelDeveloperWrite	Yes	Yes	Yes	No	No
ModelResourceOwner	Yes	Yes	Yes	Yes	Yes