Confluent Intelligence has recently announced a significant expansion of its features, introducing five new capabilities designed to enhance real-time AI systems on Confluent Cloud. One of the standout additions is the Agent-to-Agent (A2A) integration, now available in Open Preview. This capability allows Streaming Agents to seamlessly collaborate with external agents based on platforms like LangChain, CrewAI, SAP, and Salesforce, all facilitated by a Kafka backbone. Through Apache Kafka, the A2A protocol orchestrates asynchronous inter-agent communication, transforming agents built outside Confluent into event-driven entities. This setup not only fosters enhanced interaction but also ensures that every decision made by these agents is logged in real-time to system tables, which are Kafka topics equipped with structured schemas for in-depth auditability and traceability.
Another notable feature is the introduction of Multivariate Anomaly Detection, now in Early Access. This feature brings forth the ML_DETECT_ANOMALIES_ROBUST Flink SQL function, specifically designed to detect anomalies across simultaneously correlated metrics. Utilizing lightweight statistical ML techniques like ARIMA and MAD, this anomaly detection is efficient and cost-effective, particularly beneficial for IoT and operational monitoring settings since it learns in-stream without requiring batch training.
Additionally, Confluent has enhanced its Vector Search capability, which now includes support for Azure Cosmos DB and Amazon S3 vectors, reaching General Availability (GA) for Retrieval-Augmented Generation (RAG) pipelines in Flink. This development enables users to streamline their data operations across various platforms more effectively.
The platform update also sees the introduction of AWS and Azure Private Link, now in GA, facilitating secure VPC-to-VPC connectivity for model inference and external data access. This ensures that data movement and processing can occur with heightened security and efficiency.
Lastly, the update extends Confluent’s offerings with official vendor-backed support for the open-source Model Context Protocol (MCP) server. This support empowers AI agents to access Confluent Cloud data and operations seamlessly.
Not to be overlooked is the introduction of Queues for Kafka (KIP-932), which broadens Kafka’s traditional publish/subscribe model to accommodate message queue semantics within the same topic. This reduces infrastructure complexity, offering developers the flexibility to manage both paradigms within a singular system.

Collections

Confluent Intelligence has expanded with five new capabilities for real-time AI on Confluent Cloud. Agent-to-Agent (A2A) integration (Open Preview) enables Streaming Agents to collaborate with external agents from LangChain, CrewAI, SAP, and Salesforce via a Kafka backbone with full auditability. Multivariate Anomaly Detection (Early Access) introduces the ML_DETECT_ANOMALIES_ROBUST Flink SQL function using ARIMA and MAD for in-stream anomaly detection without batch training. Vector Search now supports Azure Cosmos DB and Amazon S3 for RAG pipelines in Flink (GA). AWS and Azure Private Link (GA) enable secure VPC-to-VPC connectivity for model inference. Official support for the open-source MCP server is added, and Queues for Kafka (KIP-932) extends Kafka with message queue semantics alongside its publish/subscribe model.

Confluent Intelligence Expands with New Capabilities: A2A Integration, Multivariate Anomaly Detection, and More