Natural language SQL with Ollama and DuckDB — fully local

No API key. No internet. Your questions, your data, your machine. Pipetable uses a local Ollama model to convert plain English to DuckDB SQL.

Why run everything locally?

Some data is too sensitive to send to any cloud API — even for the brief moment it takes to generate a SQL query. With Pipetable + Ollama, the entire pipeline runs on your hardware: file reading, NL-to-SQL generation, and query execution.

Install Ollama

# macOS
brew install ollama

# Linux
curl -fsSL https://ollama.com/install.sh | sh

Pull a model

ollama pull qwen2.5-coder:1.5b

qwen2.5-coder:1.5b is 986MB and runs well on CPU. For better SQL quality on a machine with enough RAM:

ollama pull qwen2.5-coder:7b   # 4.7GB, much better SQL

Start Ollama and run Pipetable

ollama serve &
pipetable ./data/

Pipetable auto-detects Ollama when no API key is set. You'll see the active provider in the startup message.

Fully local: Ollama generates SQL, DuckDB runs it, nothing leaves the machine

Ask in plain English

> show me total revenue by region
Using: sales
Thinking.....
SELECT region, SUM(revenue) AS total FROM sales GROUP BY region ORDER BY total DESC LIMIT 100

3 row(s)

region  total
─────────────
EU      179100
US       96700
APAC     18000
→ piped as _last

Chain queries with _last

Every result is saved as _last. Filter it further without re-reading the original file:

> SELECT region, total FROM _last WHERE total > 50000

Switch between providers

Pipetable auto-detects your provider in priority order:

ANTHROPIC_API_KEY  → Claude Haiku (best SQL quality)
OPENAI_API_KEY     → GPT-4o-mini (or any compatible API)
neither            → Ollama (local, no key needed)

Use OPENAI_BASE_URL to point at LM Studio, Jan, or any OpenAI-compatible local server.

More guides