AskTable + Databend: Natural Language on Cloud-Native Lakehouse SQL

AskTable supports many databases and warehouses so you can try AI analytics on the stack you already run.

We recently added Databend, a cloud-native lakehouse, alongside MySQL, PostgreSQL, ClickHouse, TiDB, and others—bringing natural-language analysis to another modern analytics engine.

From writing SQL to asking questions

As data piles up in databases, lakes, and apps, “I need a number” often means waiting on SQL, tickets, or spreadsheets. AskTable turns that into conversation: ask in plain language, get SQL and results—tables, charts, and exports.

AskTable works with leading LLMs and supports Excel, databases, warehouses, and knowledge bases—with identity, permissions, automation, and private deployment options across industries.

Today’s capabilities include:

• •Natural language to SQL with business semantics
• •Identity and permissions (e.g. “my customers last week”)
• •Many connectors—MySQL, PostgreSQL, Oracle, TiDB, Databend, ClickHouse, and more
• •Rich output—text, charts, tables, Excel
• •Automation and hooks into downstream systems
• •Self-hosted options for compliance

After SQL is generated, the target engine runs it; results may be returned directly or summarized by AI.

Databend + AskTable

Databend is an open, Rust-based lakehouse on object storage—elastic, usage-based, and often compared to an open Snowflake. It has replaced legacy warehouses and OLAP engines in many verticals.

Together you get:

• •Unified platform for storage, compute, and governance
• •Flexible SQL without hand-tuning every index for AskTable-generated queries
• •Databend Cloud pay-as-you-go to control cost

Hands-on: five questions on GitHub logs

Using Databend Cloud (https://app.databend.cn) and GitHub events from 2025-05-15, we test AskTable end to end.

Step 1: Connect the datasource

Create a session in AskTable, connect Databend Cloud, and pick tables (up to 100). AskTable ingests schema and stats for semantic modeling.

Wait for metadata sync; large schemas may take a moment before you query.

Step 2: Ask questions

Q1: What are the top 10 hottest projects by the data?

Example SQL:

The sample is single-day data; SQL time ranges follow the data’s min time to “now.”

Observability in AskTable:

Databend Cloud query logs (~640 ms execution in one run—end-to-end latency may be higher; tune as needed):

Execution plan:

Both AskTable and Databend expose observability for performance work. Private AskTable can ship with observability stacks.

Q2: Most popular languages—top 10, excluding null.

Q3: Repos with the most stars—top 10.

Q4: Who was active on Databend-related repos on May 15?

Initially empty: generated SQL used like '%Databend%' (capital D). A quick hint fixes it:

Terminology and aliases can reduce similar issues.

Q5: How many GitHub events on May 15?

Takeaways

• •Strong accuracy on single-table structured queries.
• •RAG-style hints help interactive self-service.
• •Field-level hints on github_events produced precise SQL—LLMs are powerful here.

Best experience: embed AskTable inside your product so teams get AI analytics where they already work.

---

Republished from the Databend WeChat article.