Starburst Data Inc., maker of a distributed query engine built on the open-source Trino project, today introduced what it calls the industry’s first artificial intelligence-ready data platform designed for the “agentic workforce.”

That’s an organizational model in which humans and AI agents collaborate to make decisions while maintaining governance and compliance across distributed data stores.

Announced at the company’s AI & Datanova event, the new platform integrates model-to-data architectures, multi-agent interoperability and an open vector store built on the open-source Apache Iceberg table format. Starburst said its goal is to give enterprises a way to unify data products and metadata in a single governed environment, allowing AI agents to reason and act securely without the need to move data between systems.

Starburst’s approach contrasts with more common AI architectures that centralize data in a single location. That requires organizations to move or copy data between platforms, a task that adds time and cost and may introduce security vulnerabilities.

The model-to-data framework allows AI models to access governed data across on-premises or cloud environments without breaching security or privacy rules, said Nathan Vega, Starburst’s senior director of product marketing. That addresses a key challenge for enterprises trying to put AI into operation at large scale.

“Most of the customers we talk to are being held back by fragmented data and the lack of governance,” Vega said. “Our platform is built around federated analytics and governance, so they can use structured and unstructured data in one place and accelerate any project that requires data, whether that’s [business intelligence], analytics or AI.”

Cost controls

Starburst said its new release also adds model usage monitoring and control. “We support having users entitled to specific models and can track prompts, token-level usage and query volumes,” Vega said. That allows information technology organizations to control costs and monitor model performance across workloads.

Starburst co-founder and Vice President of AI Matthew Fuller said usage monitoring extends beyond simple token counts. “It could be tokens in and out or finer-grained metrics like reasoning tokens,” he said. “In the future, we want to let customers specify limits based on dollar amounts rather than just token counts.”

A central feature of the release is support for the emerging Model Context Protocol, which enables interoperability between multiple AI agents. “We think there’s a future where enterprises have hundreds or thousands of agents running the business,” Vega said. “The MCP layer becomes the standardized access point for those agents to get tools, fetch data products or even open ServiceNow tickets. It becomes the control and governance layer for agentic workflows.”

Fuller said Starburst provides its own MCP server and agent application programming interface, allowing users to create, manage and orchestrate multiple agents alongside the Starburst agent. “This lets enterprises develop multi-agent applications that complete tasks of growing complexity,” he said.

Regulation-friendly

Starburst said its federated architecture is also designed to maintain data sovereignty across jurisdictions. “We have the ability to run a cluster in the [European Union]and other clusters somewhere else,” Fuller said. “When you’re querying, the acceptable data can transfer across regions, but data that needs to stay put can be preprocessed or aggregated locally,” to comply with laws such as the European Union’s General Data Protection Regulation.

Fuller acknowledged that compliance requires design and implementation by customers, but said Starburst’s architecture provides the necessary foundation. “It’s not a turnkey thing,” he said. “The features enable customers to enforce compliance, but there’s definitely thought and design required.”

The new release also unifies access to vector stores, specialized systems that store, index and quickly retrieve numerical representations of unstructured data. That enables retrieval-augmented generation and search tasks across Iceberg, PostgreSQL with pgvector, Elasticsearch and other systems. “If it’s in Elastic, it stays in Elastic,” Fuller said. “But if customers want to keep vector embeddings on the lake, they can store them directly in Iceberg and run semantic search over those embeddings using Starburst’s engine.”

The platform also includes new visualization tools. AI agents can present data as charts and graphs, providing what Vega described as “a natural language response, a visualization or a combination of both.”

Starburst emphasized that its lakehouse platform avoids locking in customers by supporting open data standards. Vega said the company aims to offer flexibility “whether agents are on-prem, in one cloud or another cloud,” without centralizing data on a single platform.

The new features will be generally available in the fourth quarter of 2025.

Image: SiliconANGLE/Microsoft Designer