Databend

Enterprise Data Warehouse for AI Agents

one warehouse for all your data

SQL analytics, vector search, and full-text search in a single Rust warehouse on object storage: fast, elastic, and grows with your data — with sandboxed Python UDFs for agents.

💡 Why Databend?

Databend is an open-source data warehouse built in Rust. Analytics, search, and agent workloads run on the same tables, so there is no second system to sync.

📊 One Warehouse SQL analytics, vector search, full-text search, auto schema evolution, transactions.	🤖 Agent-Ready Sandboxed Python UDFs for agent logic, SQL for orchestration.
🏢 Enterprise Scale Fast vectorized engine, elastic compute on object storage: S3, Azure, GCS.	🌿 Branching Git-like data versioning. Agents experiment safely on production snapshots.

📈 Performance

Published results on Databend Cloud with default settings. Each page includes the SQL to reproduce it.

Benchmark	Result
TPC-H SF100 vs. Snowflake	1.25× faster, ~60% lower cost
Data ingestion vs. Snowflake	5.15× faster, 91% lower cost
Vector & full-text search	10M docs: vector p50 41 ms, full-text p50 35 ms
TPC-H SF1000 (1 TB) scaling	4.1× faster on Large vs. Small warehouse

See all benchmarks.

⚡ Quick Start

Cloud (Recommended) — Sign up, create a warehouse, and run SQL in the worksheet. Cloud guide →

Self-hosted — Run the all-in-one image and connect with BendSQL:

docker run -d -p 127.0.0.1:8000:8000 datafuselabs/databend
bendsql -u root

See Docker image options and self-hosted deployment for credentials, storage, and production setup.

Your First Query

Analytics, full-text search, and vector search in one table:

CREATE TABLE docs (
    id INT,
    content STRING,
    embedding VECTOR(3),
    INVERTED INDEX content_idx(content),
    VECTOR INDEX embedding_idx(embedding) distance = 'cosine'
);

INSERT INTO docs VALUES
    (1, 'SQL analytics for agents', [1.0, 0.0, 0.0]),
    (2, 'Vector search for documents', [0.0, 1.0, 0.0]);

SELECT id, content FROM docs WHERE MATCH(content, 'analytics');

SELECT id, content FROM docs
ORDER BY COSINE_DISTANCE(embedding, [1.0, 0.0, 0.0]::VECTOR(3))
LIMIT 1;

Both queries return document 1. Learn more: Vector search · JSON & search.

🤖 Agent-Ready Architecture

Databend's Sandbox UDF enables flexible agent orchestration with a three-layer architecture:

• Control Plane: Resource scheduling, permission validation, sandbox lifecycle management
• Execution Plane (Databend): SQL orchestration, issues requests via Arrow Flight
• Compute Plane (Sandbox Workers): Isolated sandboxes running your agent logic

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

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

🚀 Use Cases

• AI Agents: Sandbox UDF + SQL orchestration + branching for safe operations
• Analytics & BI: Large-scale SQL analytics — Learn more
• Search & RAG: Vector + full-text search — Learn more

🤝 Community & Support

• 📖 Documentation
• 💬 Join Slack
• 🐛 Issue Tracker
• 🗺️ Roadmap
• 🛠️ Contributing — AI-assisted PRs are welcome; a responsible human must review the full diff

Contributors are immortalized in the system.contributors table 🏆

📄 License

Databend is licensed under Apache 2.0. A small set of enterprise features is under Elastic 2.0, see LICENSING.md | Licensing FAQ

---