What it is

Databend open-source cloud-native data warehouse built in Rust. It targets enterprise AI workloads, SQL analytics, vector search, full-text search, Python UDF sandbox, lakehouse workloads on S3, Azure, GCS. Project lives in database and infrastructure ecosystem, with topics covering OLAP, bigdata, serverless, geospatial, Elasticsearch, Snowflake.

Problem: teams often split analytics, search, agent logic, and sandboxed Python across separate systems. Databend puts these functions in one engine. It uses SQL for orchestration, sandboxed UDFs for agent logic, transactions for reliability, branching for production snapshots. It also supports auto schema evolution and elastic compute.

Key capabilities

• SQL analytics engine for large-scale queries, OLAP workloads, and BI-style reporting.
• Vector search and full-text search for retrieval, RAG, and document-style query workflows.
• Python sandbox UDFs let users define agent logic, LLM calls, tool use, and reasoning inside database functions.
• SQL orchestration over sandbox workers uses Arrow Flight between Databend and compute plane.
• Branching provides Git-like data versioning, allowing agents to operate on snapshots without changing production tables.
• Transactions and auto schema evolution support stateful analytics and changing data shapes.
• Object-storage backed lakehouse architecture supports S3, Azure, and GCS, with elastic compute.

Who uses it and how

• AI agent teams use sandbox UDFs and SQL to run agent logic close to warehouse data, with branching for safe experimentation.
• Analytics teams use SQL for large-scale reports and BI queries over lakehouse tables stored in object storage.
• Search and RAG builders use vector plus full-text search to retrieve rows for prompts, documents, or product data.
• Developers test locally with Python databend-driver or run full warehouse with Docker image datafuselabs/databend.
• Teams evaluating Snowflake or Elasticsearch can compare Databend as one open-source engine for analytics, search, and agent workflows.

Getting started

Typical paths are Databend Cloud, local Python via pip install "databend-driver[local]>=0.34.0" with Python 3.12 or 3.13, or Docker with docker run -p 8000:8000 datafuselabs/databend.

When to use it — and when not to

Use Databend when analytics, vector search, full-text search, and Python agent UDFs need one SQL engine over object storage. It fits self-hosted or cloud lakehouse workloads where branching, transactions, and sandboxed UDFs matter, but self-hosters must operate object storage, compute, and sandbox workers. Avoid it when mature single-purpose search service, fully managed warehouse, or strictly permissive license is required; repository is young, lists Apache 2.0 plus Elastic 2.0, has 565 open issues, and shows 0 contributors in provided metadata, so evaluation should check current governance and operational maturity.

pip install "databend-driver[local]>=0.34.0"

from databend_driver import connect

conn = connect("databend+local:///./local-state")
print(conn.query_row("SELECT 'Hello, Databend!'").values())

docker run -p 8000:8000 datafuselabs/databend

-- Define your agent logic
CREATE FUNCTION my_agent(input STRING) RETURNS STRING
LANGUAGE python HANDLER = 'run'
AS $$
def run(input):
    # Your agent logic: LLM calls, tool use, reasoning...
    return response
$$;

-- Orchestrate agents with SQL
SELECT my_agent(question) FROM tasks;