Starburst pushes lakehouse boundaries with multi-agent AI and unified vector search

Starburst is updating its AI services to help enterprises build multi-agent workflows and monitor these agents for compliance and cost effectiveness.

The data lakehouse provider, earlier in May, released a set of AI services, comprising the Starburst AI Agent and AI Workflows, across the Starburst Enterprise Platform and Starburst Galaxy, aimed at adding an agent-supporting layer to its lakehouse platform.

To enable enterprises to build multi-agent workflows on its platform, Starburst has added a new MCP server and agent API.

Industry players are racing to offer MCP servers to reduce enterprises’ burden for custom integrations, especially while setting up multi-agent workflows.

While Databricks supports MCP through managed servers, external connections, and custom hosting, Snowflake offers a managed MCP server, currently in preview, and open MCP projects.

Additional governance features for control and cost

To help enterprises manage agents more effectively and track costs associated with them, Starburst is adding dashboards that will allow teams to track, audit, and control AI usage.

Governance of agents, according to analysts, is paramount as agents chain tools and data autonomously, which could result in compliance issues.

HyperFRAME Research’s practice leader Stephanie Walter points out that vendors should also allow enterprises to trace agent reasoning and actions and set rate limits or kill switches.

“As AI agent deployment and usage peak, I expect agent governance to be a real headache for enterprises. Think about how difficult it will be to ensure thousands of AI agents perform safely, ethically, and compliantly. A solid and easy-to-use agent governance solution will be a differentiator,” Walter said.

However, Starburst rivals, such as Databricks and Snowflake, offer agent governance features with Unity Catalog and Horizon, respectively.

Analysts point out that the agent usage tracking features added to Startburst are not only important for controlling over-expenditure but are also becoming table stakes across data lakehouses.

While Databricks’ Mosaic AI Gateway offers rate limits, usage tracking, and inference logs and tables, with cost monitoring through system tables, Snowflake’s Cortex AI Observability offers tracing, evaluations, and cost and usage analysis, along with community-built cost dashboards.

Unified access to vector stores

Another update to Starburst AI services is the lakehouse provider adding vector search and unified access to vector stores, enabling retrieval augmented generation (RAG) and search tasks across Iceberg, pgvector, and Elasticsearch.

While vector search in itself as a feature is not new, unified access to multiple vector stores is what sets Starburst apart from rivals, analysts say. The updates, which will have their own additional costs, are expected to be made generally available by the end of this year, the data lakehouse provider said without providing details on pricing.