A warehouse rebuilt so AI agents can live inside SQL

What it does

Databend is an open-source data warehouse rebuilt in Rust that bundles analytics, vector search, full-text search, and auto schema evolution into one engine backed by object storage. It runs AI agent logic natively via sandboxed Python UDFs defined in SQL, orchestrating agents alongside traditional queries with transactions and git-like branching for safe experimentation on production data.

The interesting bit

Instead of piping warehouse data out to external agent frameworks, Databend hosts isolated sandbox workers inside a three-layer architecture: a control plane handles permissions and lifecycle, the execution plane issues SQL via Arrow Flight, and the compute plane runs your Python agent logic. The pitch is that agents query, reason, and act without leaving the warehouse’s permission model.

Key highlights

• Unified engine for analytics, vector search, and full-text search with auto schema evolution
• Sandboxed Python UDFs orchestrated through standard SQL for agent logic
• Git-like branching lets agents experiment on production snapshots without breaking live data
• Cloud-native elastic compute targeting S3, Azure Blob, and GCS
• Dual-licensed under Apache 2.0 and Elastic 2.0

Verdict

Worth evaluating if you need one backend for both large-scale BI and AI agent orchestration. Probably overkill if you just want a standalone vector database or a lightweight SQL analytics replica.